fix(resolve): map grok-build to grok in the pid walk

[apstndb]

Fixes #859.

_agmsg_agent_binaries has no grok-build arm, so the type falls through to claude codex gemini and grok-build watchers are not liveness-gated.

Independently mergeable. Split out of #856 because cursor monitor only needs the cursor-agent arm. Reading detect_proc from the manifest stays #631.

Do not list agent (Homebrew grok-build and the Cursor CLI installer share that basename). The policy is on #856; this PR comments the omission.

#858 stays a draft until this and #857 have landed.

Test plan

• bats tests/test_resolve_project.bats --filter agent-binaries
• bats tests/test_resolve_project.bats
• CI green

[fujibee]

Thanks for this, and sorry it sat.

This was red for a reason that had nothing to do with your change. One macOS
shard was stopped at exactly 25 minutes — this workflow's job cap — and GitHub
records a job that hits the cap as cancelled. The aggregate bats check then
went red naming no shard and no test, which from the outside is indistinguishable
from the change being rejected.

Every ubuntu shard, every Windows leg and the other three macOS shards were
green. Re-running the stopped job passed with nothing else changed, so the run
here is green now.

The cap is being hit because a handful of tests are slow: eight cases account for
about 30% of the suite's runtime, seven of them in sync start. Work is open on
that (#876), on the queueing that makes a slow shard slower (#877), and on making
this exact failure legible instead of a bare result: cancelled (#878).

Nothing is needed from you. Review to follow.