diagnostics: add startup phase trace for runtime bootstrap

[sonukapoor]

Summary

Add an opt-in startup phase trace for runtime bootstrap diagnosis.

This PR adds lightweight, environment-gated tracing in src/edge_runtime.cc to emit per-phase startup timing markers during RunScriptWithGlobals.

• env var: EDGE_STARTUP_TRACE=1
• output channel: stderr
• output shape: JSON-like one-line entries with:
  • edge_startup_trace (phase name)
  • delta_ms (time since previous marker)
  • total_ms (time since traced startup begin)

Why

Current benchmark work shows startup-sensitive gaps, but wall-clock totals alone do not identify which bootstrap phase dominates.

This trace is intended as a diagnosis artifact so follow-up performance changes can target concrete hotspots and produce before/after evidence per phase.

Scope

• Diagnostics only.
• No behavior or compatibility changes when EDGE_STARTUP_TRACE is unset.
• No benchmark workload changes in this PR.

Example usage

EDGE_STARTUP_TRACE=1 ./build-edge/edge benchmarks/workloads/empty-startup.js
EDGE_STARTUP_TRACE=1 ./build-edge/edge -e ""

Example output (local)

{"edge_startup_trace":"install-process-object","delta_ms":2.51975,"total_ms":2.81742}
{"edge_startup_trace":"bootstrap.per_context.primordials","delta_ms":9.84379,"total_ms":12.7747}
{"edge_startup_trace":"bootstrap.realm","delta_ms":4.55288,"total_ms":17.7587}
{"edge_startup_trace":"bootstrap.node-and-web","delta_ms":16.6824,"total_ms":34.4411}

This local sample suggests bootstrap.node-and-web is a major hotspot candidate for the next narrow cold-start optimization pass.

[sonukapoor]

Quick cold-start update:

Using EDGE_STARTUP_TRACE, I tested a few narrow startup optimizations with A/B runs against the same baseline binary.

Tried:

• lazy-load in internal/bootstrap/switches/is_main_thread.js
• lazy wiring in internal/bootstrap/node.js (source-map/builtin-module helper path)

Outcome:

• no stable speedup yet
• one attempt was noise-level
• one regressed by ~1% on repeated runs
• both were reverted

Current status: diagnostics improved, but no confirmed startup win yet.

Next: split bootstrap.node into finer sub-phases, then target the heaviest sub-phase with a narrow follow-up and report only if the gain is repeatable.