Add tracelite profiling hooks and production gate

[danReynolds]

Summary

• Adds a profile-only tracelite bridge behind RESQLITE_PROFILE + RESQLITE_TRACELITE, with no runtime attach unless a trace region is configured.
• Registers resqlite span/counter metadata so tracelite reports show stable names for database, worker, stream, dispatch, decode, diagnostics, and fanout signals.
• Adds trace_sqlite native-asset hook support plus a CI smoke fixture that verifies the embedded sqlite3mc shim/runtime path without requiring a sibling tracelite checkout.
• Adds benchmark/profile/run_tracelite_profile.dart, the parity workflow that creates a tracelite region, runs the legacy profile harness with tracing enabled, exports workload-summary and graph data, validates the graph-data bundle, and writes a parity diff against the legacy profile JSON.
• Adds top-level benchmark/run_tracelite.dart, the resqlite pre-publish gate over tracelite suite-history, scoped to the calibrated measured-elapsed release gate and optional dashboard graph-data export.
• Teaches the benchmark dashboard to consume tracelite graph data when present while keeping raw trace regions and legacy profile JSON out of docs/.
• Keeps the release path unchanged: tracing is profile-mode, best-effort, and disabled unless both compile-time flags and TRACELITE_REGION are present.

Notes

• The wrapper shells out to a local tracelite checkout instead of adding tracelite as a resqlite package dependency.
• The legacy profile JSON remains as a compatibility/parity bridge. New GitHub Pages profiling views should read tracelite graph-data bundles.
• point-select and keyed-pk-subscriptions remain diagnostic for the release gate until their current variance fits below the 50% threshold ceiling.

Validation

• /Users/dan/Coding/flutter_arm64/bin/dart pub get
• /Users/dan/Coding/flutter_arm64/bin/dart run build_runner build --delete-conflicting-outputs
• /Users/dan/Coding/flutter_arm64/bin/dart analyze
• /Users/dan/Coding/flutter_arm64/bin/dart test --timeout 60s
• /Users/dan/Coding/flutter_arm64/bin/dart test test/tracelite_profile_test.dart test/tracelite_profile_workflow_test.dart test/tracelite_benchmark_workflow_test.dart
• /Users/dan/Coding/flutter_arm64/bin/dart run tool/trace_sqlite_smoke.dart
• /Users/dan/Coding/flutter_arm64/bin/dart run benchmark/check_generated_data.dart
• /Users/dan/Coding/flutter_arm64/bin/dart run benchmark/check_experiment_signals.dart
• node -e "const fs=require('fs');const html=fs.readFileSync('docs/benchmarks/index.html','utf8');const scripts=[...html.matchAll(/<script>([\\s\\S]*?)<\\/script>/g)].map(m=>m[1]).join('\\n');new Function(scripts);console.log('dashboard script syntax ok')"
• /Users/dan/Coding/flutter_arm64/bin/dart run benchmark/profile/run_tracelite_profile.dart --tracelite-root=/Users/dan/Coding/tracelite --dart=/Users/dan/Coding/flutter_arm64/bin/dart --label=codex-production-readiness-smoke --out-dir=/tmp/resqlite-tracelite-profile-production-readiness --graph-data-dir=/tmp/resqlite-tracelite-profile-production-pages
• Profile smoke parity: parity-diff.txt reported 0 deltas for workload timings, memory deltas, SQLite diagnostics, and noop floors between legacy profile JSON and tracelite workload summary.
• Profile graph-data smoke: tracelite validate-graph-data passed; index.json reported 4 workload_summary rows, 41 workload_operations rows, and 132 workload_memory rows.
• /Users/dan/Coding/flutter_arm64/bin/dart benchmark/run_tracelite.dart --tracelite-root=/Users/dan/Coding/tracelite --dart=/Users/dan/Coding/flutter_arm64/bin/dart --label=codex-gate-smoke --out-dir=/tmp/resqlite-tracelite-gate-smoke-real --graph-data-dir=/tmp/resqlite-tracelite-gate-smoke-pages --runs=1 --interfaces=resqlite --policy-peers=resqlite --policy-scenarios=narrow-batch-insert --policy-metric=measured_elapsed_ns --min-repetitions=1 --max-repetitions=2 --target-rse-percent=100 --threshold-floor-percent=5 --threshold-ceiling-percent=100 --noise-gate-floor-percent=5 --noise-gate-ceiling-percent=100 --no-strict
• Gate smoke result: suite-history completed with run status ok, policy status ready, and graph-data export wrote 1400 scenario_series rows plus 10 peer_summary rows.

[danReynolds]

Updated this PR with the second parity slice for tracelite-as-profile-replacement:

• Added resqlite.profile.workload scopes to benchmark/run_profile.dart so tracelite reports now group noop, single_insert, point_query, and merge_rounds with the old sample counts.
• Added correlated SQLite diagnostics gauges around each profile workload: page cache, schema, stmt heap, WAL bytes, stream count, and reader-busy.
• Added workload scopes to benchmark/profile/many_streams_writer_profile.dart so the stream profile report groups baseline, disjoint, and overlap and includes stream invalidation/intersection/dispatcher counters.
• Cached interned tracelite strings in the resqlite bridge; the parity run exposed string-pool overflow from repeated SQL/name interning before this fix.

Local validation:

• /Users/dan/Coding/flutter_arm64/bin/dart analyze lib/src/tracelite_profile.dart benchmark/run_profile.dart benchmark/profile/many_streams_writer_profile.dart test/tracelite_profile_test.dart
• /Users/dan/Coding/flutter_arm64/bin/dart test test/tracelite_profile_test.dart
• tracelite full dart analyze
• tracelite full dart test
• tracelite production suite across sqlite3,drift,sqlite_async,resqlite
• side-by-side old benchmark/run_profile.dart vs tracelite-enabled PR run
• tracelite-enabled many_streams_writer_profile.dart

CI for head c581acd is green: Tests, generated benchmark freshness, and raw-profile-JSON guard all passed.

Remaining migration blocker is not raw signal capture anymore. It is export/workflow compatibility: tracelite still needs a workload-summary JSON/report path for old fields like noop-floor subtraction, work_us, RSS deltas, and many-streams fanout-delta tables before we delete the old profile JSON/diff machinery.

[danReynolds]

Parity update after a519408:

• Added resqlite profile parity signals for tracelite: exact profile samples, RSS before/after/peak, profile counter snapshots, and many-streams fanout counters.
• Added tracelite workload-summary export locally in tracelite commit cb53883. That exporter emits old-compatible workload JSON plus markdown from a .tlt-region.
• Validated a fresh traced benchmark/run_profile.dart run against tracelite workload-summary; existing benchmark/profile/diff.dart accepted the tracelite summary JSON and reported +0 deltas for p50/p90/p99/max/work_us, RSS delta, SQLite diagnostic deltas, noop floors, and profile counter deltas.
• Validated many-streams fanout medians from tracelite summary against the old raw many-streams JSON: baseline/disjoint/overlap matched exactly for writer/yield/total/invalidate/intersection timings and intersection entries.
• Local checks: resqlite focused analyze and dart test test/tracelite_profile_test.dart pass; tracelite full dart test passes when resqlite is overridden to this PR worktree.
• GitHub CI on this PR head passed: Guard against raw profile JSONs, Tests, and Benchmark generated data freshness. Benchmarks (main) skipped as expected on this PR.

This closes the parity gap I found earlier: remaining follow-up is migration cleanup, not signal/export discovery.