tooling: add profiling harness for V8 cpu/heap/tick profiles

[7rulnik]

Summary

Adds a small profiling harness so the next round of perf work can be driven by measured hot frames instead of guesses. Pairs naturally with the existing scripts/bench-diff.sh workflow — bench-diff tells you whether a change is a win, this harness tells you where to look for one.

• scripts/profile-driver.js — runs ONE scenario (init / opf / gpf / eachmap-gen / eachmap-orig) against ONE fixture, sized so a single node invocation produces a useful profile. Warms up before the measured loop so V8 has tier-upped before sampling starts. Honors SCENARIO / FIXTURE / ITERS / WARMUP env vars.
• scripts/profile.sh — wraps node with --cpu-prof, --heap-prof, and --prof flags and writes outputs to tmp/profiles/. Honors a MODE env var (all / cpu / heap / prof) so individual modes can be run alone.
• scripts/analyze-cpuprofile.js — bucket-by-frame self-time summary of a .cpuprofile (top N).
• scripts/analyze-heapprofile.js — bucket-by-frame self-allocation-bytes summary of a .heapprofile (top N).
• .gitignore — adds tmp/ so profile output stays out of version control.

The CPU profile output (via Chrome DevTools or the analyzer) was what surfaced the gpf hot frame addressed in #68.

Usage

# default: opf scenario on babel.min.js.map, all three modes
scripts/profile.sh

# specific scenario / fixture
SCENARIO=gpf FIXTURE=vscode.map scripts/profile.sh

# cpu only
SCENARIO=init MODE=cpu scripts/profile.sh

# read top frames from the resulting profile
node scripts/analyze-cpuprofile.js tmp/profiles/gpf.cpuprofile 15

Test plan

• Each scenario runs to completion under each mode against babel.min.js.map
• Output files (.cpuprofile, .heapprofile, .prof.txt) are produced in tmp/profiles/
• Both analyzers produce sane top-N tables
• No production code is touched