Profile TypeScript execution and reduce module-resolution probes (GOL-347, GOL-350)

[noise64]

• resolves https://linear.app/golem-cloud/issue/GOL-347
• resolves https://linear.app/golem-cloud/issue/GOL-350
• adds feature-gated execution phases and bounded module/filesystem counters for fresh-runtime TypeScript compiler profiling
• replaces ordinary path-based readFileSync chunk loops with one private native whole-file read while preserving descriptor and custom-flag fallbacks
• adds a positive-only probe session scoped to each outermost CommonJS execution, covering file and path-classification probes while preserving runtime-local cwd semantics and filesystem-mutation invalidation
• publishes fresh five-sample P2/P3 TypeScript profiles from the combined source and balanced optimized Ajv comparisons with five alternating executions per mode; disabling the session produces 428 = 392 system + 0 outer-session + 36 invocation-local, while enabling it produces 428 = 322 system + 70 outer-session + 36 invocation-local
• removes 17.9% of physical metadata probes in both targets; optimized P2 improves 0.79% at the median and 2.67% in throughput, while P3 improves 1.33% and 1.77%; the five-sample timings are directional rather than a durable latency claim
• retains GOL-348 as a bounded cache-lifetime/invalidation investigation and GOL-349 after GOL-418

Validation includes schema and exact-source currentness checks for all four changed reports, the module-loader architecture contract, focused and broad P2/P3 module-resolution and filesystem runtime coverage, affected P2/P3 readfile, read_file, module_stat, rmdir, and unlink Node-compat candidates, the 10-lane production/profiling skeleton Clippy matrix, formatting, and independent whole-diff and correction-delta reviews.

[noise64]

Exact candidate: ca96cac7800089062826532ebb337f57554bec35.

Correction-delta review is closed:

• Codex: clear; all four original findings resolved.
• Claude: clear; all ten correction obligations confirmed. Four low-severity, non-blocking follow-up observations were recorded (feature-label deduplication, two report mutation guards, an uncertain UTF-8 error-path nuance, and a pre-existing custom-flag directory-read divergence).

Independent exact-SHA validation passed:

• formatting and paired P2/P3 report currentness
• exact Agentic TypeScript and fs DTS tests
• P2/P3 focused fs_read_file_sync_fast_path
• P2/P3 17-case Node readfile family and dedicated readFileSync filters
• development-profile semantics
• P2/P3 default and maximal failure-safe skeleton Clippy lanes

The tracked tree remained clean and the skeleton manifest/artifact cleanup contract was restored after validation.

[noise64]

CI classification for exact head ca96cac7800089062826532ebb337f57554bec35:

• build passed and the full matrix was admitted.
• skeleton-clippy failed only at the stacked base's two known Rust 1.98 clippy::chunks_exact_to_as_chunks diagnostics in src/builtin/web_crypto.rs:680 and :700.
• The failure is outside the GOL-347 diff and matches the independently recorded GOL-421 authorization gate; no GOL-347 diagnostic appeared.

Remaining exact-SHA CI jobs continue to run. The branch will not absorb a third-cycle GOL-421 correction without the required authorization checkpoint.