fix(measure): report why a server died instead of waiting out the timeout

[VickyXAI]

Found while verifying the published 0.45.1 end-to-end.

The bug

Pointing scripts/measure-tool-schema.mjs at a command that fails to start gave you a silent 60-second wait and then timed out — for what was sh: blockrun-mcp: command not found in the first 200ms.

We advertise this script for measuring other people's servers, where a mistyped command or a package that 404s is the common case. A tool whose most likely failure takes a minute and then blames the wrong thing is one people try once.

The fix

• stderr is piped, not ignored (tail only, so a chatty server isn't buffered whole).
• A child that exits before the handshake rejects immediately, carrying its exit code and its own error text.
• Timeout raised 60s → 120s, and the message now says what to do about it — a cold npx install of a package this size genuinely exceeds a minute, which is how the bad message got found in the first place.

Failure goes from 60s and a wrong reason to 0.25s and the right one:

Error: node /nonexistent-server.js exited with code 1 before completing the handshake.
Error: Cannot find module '/nonexistent-server.js'

Also worth recording

npx -y @blockrun/mcp@<version> run from inside this repo resolves to the local checkout and dies with exit 127. Measure published packages from a neutral directory. Done that way, 0.45.1 measures 12,900 — matching the card exactly.

Verification

• 431/431 tests pass (+1: asserts the failure names the exit, carries the child's stderr, and returns in under 10s rather than on the timeout).
• Typecheck and sync-brand-numbers --check clean; happy path unchanged at 12,900.

No version bump: scripts/ is not in the npm tarball, so this changes nothing for installed users and does not need a release.

🤖 Generated with Claude Code

https://claude.ai/code/session_01WUL3ExR4Nz7uebxKaKwjDi