instrument: the lock-holder and scheduled-job instruments, a notice told once its send lands, and the check command under test

[ohdearquant]

Closes #52, closes #53, closes #51.

Two instruments join the set built from one config: a lock holder, which reads who holds each named lock through lsof, with the agent's own pid file as the control, and reports a lock lsof cannot stat as unreadable rather than unheld (the status error on <path>: <why> line is read; a missing file stays "no holder"); and a scheduled job, which reads each job's artifact age against its bound first and the supervisor's last exit second, reporting stale and failed. A dead-instrument notice marks its recipient told only once the send lands; a send that fails leaves the crossing untold, so the next pass tells it again. lion agent --check has a test that pins the measurement it prints and the sentence saying it is unvetted.

ADR-0014 S9 and S11 state what holds, with the tests named; ADR-0015 and the index follow.

Suite: 100% coverage on 3.11, 3.12 and 3.14; pre-commit clean.

[ohdearquant]

4059d74: six review findings fixed. The lock-holder read a name lsof escapes (a tab, a newline, a backslash, a byte outside the locale, on its n lines and its status-error lines) as no holder; a name no configured path resolves to now fails the control and is named as unmapped, and the real-lsof test holds a lock with a tab in its name. A scheduled job refuses a max_age_h that is a bool, not a number or NaN at build (NaN read every artifact fresh); its artifact is read with one stat and anything but a regular file is a finding; a launchctl read that failed or did not finish is unread, and not loaded is only what launchd says. ADR-0014 C4 now agrees with S11 (a recipient is marked told once its send lands); ADR-0015 cites the hand-run check's test and lists lsof. Gates: pre-commit twice, pytest with coverage on 3.11, 3.12 and 3.14 at 100%.