benchmarks: clarify startup profile capture behavior on main

[sonukapoor]

Clarify and harden benchmark startup-profile capture/reporting so the output is accurate on current main.

This PR updates:

• benchmarks/capture-edge-startup-profile.mjs
• benchmarks/README.md

Changes

1) Capture script behavior is now accurate and less brittle

• Parse potential startup-profile JSON from both stdout and stderr.

• Detect profile-like JSON more defensively instead of assuming a specific key shape only.

• When no startup profile JSON is emitted, write a neutral, accurate reason:

  • before: binary built without EDGE_ENABLE_STARTUP_PROFILE
  • now: EDGE_STARTUP_PROFILE did not emit startup profile JSON on this checkout

• Markdown output now states that profiling output was not detected for the run, rather than asserting a specific compile-time cause.

2) README wording updated to match real harness behavior

• Document that the harness attempts startup profile capture by setting EDGE_STARTUP_PROFILE=1.
• Add explicit verification command:
  • EDGE_STARTUP_PROFILE=1 ./build-edge/edge -e ""
• Keep the rebuild attempt with -DEDGE_ENABLE_STARTUP_PROFILE=ON as a possible path, but avoid implying it is always sufficient on every checkout.

Reason

A prior issue checkpoint called out that startup-profile outputs remained unavailable even after a rebuild attempt with -DEDGE_ENABLE_STARTUP_PROFILE=ON.
This PR avoids over-claiming the cause and makes artifacts/reporting accurate for diagnosis work.

Verification

./benchmarks/run.sh empty-startup
cat benchmarks/results/empty-startup.edge-profile.json
cat benchmarks/results/empty-startup.edge-profile.md