test(coverage): drive sdk-go to >=95% coverage

[mastermanas805]

Summary

• Refactor examples/agent-bootstrap/main.go + examples/provision-all/main.go to extract a testable Run(ctx, client) error function; main() stays as a thin wrapper.
• Add main_test.go for each example with an httptest-backed mock api server covering happy + error branches.
• Add instant/coverage_test.go covering every error-return branch in deploy.go (5xx, decode failure, transport error, tarball read failure, nil writer, SSE write error), storage.go, cache.go, mongodb.go, queue.go, webhook.go, resources.go, and claim.go.

Coverage

• Before: 59.4% (instant 86.4%, examples 0%)
• After: 97.8%
  • examples/agent-bootstrap 98.9%
  • examples/provision-all 98.6%
  • instant/ 97.3%

Per user mandate "95% across all repos without any exception" — no codecov ignores; the examples are actually exercised end-to-end against an httptest server.

Test plan

• go test ./... -short -coverprofile=coverage.out -covermode=atomic — green, 97.8% total
• go test ./... -race -count=1 -short — green
• go vet ./... + go build ./... — green
• Exported SDK API unchanged (refactor only extracts internals from main; signatures of instant.Client methods untouched)

Refactor rationale: extracting Run(ctx, client, ...) from main() is the industry-standard pattern for testing executable examples — not coverage theatre. The new Run function can be called from a documentation example, a sibling tool, or a different runtime without changing behaviour.

Co-Authored-By: Claude Opus 4.7 (1M context) noreply@anthropic.com