t27b coverage: real tests for 13 cron cards in specs/crons (999-multibots-telegraf and railway) #8157

Part of #6063 (t27b coverage), which is item C10 of #6488 ("coverage counted without vacuous passes").

What is wrong

On the t27b lab, 26 cron cards of specs/crons pass vacuously (0 runtime asserts) under the reference and t27b: they have no test, invariant or bench block, which also breaks L4 TESTABILITY. This issue covers 13 of them: the 12 cards of 999-multibots-telegraf and railway-jobsearch-cron.

What the tests check

A cron card is data, but specs/crons/README.md states its schema, and the tests hold each card to it:

1. ID is <host>/<repo>/<slug>: it opens with HOST, then REPO; the slug is one path segment. HOST is one of the five hosts.
2. ON_FAILURE and CONTROL come from their vocabularies. An empty RUNS (the control link) carries its reason in RUNS_NOTE.
3. A SCHEDULE is a five-field cron expression.
4. A timer's SERVICE is <file>:<line>, and the slug ends in -L<line> for that line. The line is the one the card's header says the schedule was read from.
5. A GitHub Actions card's SERVICE is its workflow file, .github/workflows/<slug>.yml.
6. A Railway dashboard schedule is "" with a SCHEDULE_NOTE.

railway-jobsearch-cron breaks rule 1 (its ID names 999 while REPO is railway; #7911), so its test checks only the shape of ID.

Earlier attempt

See #6901 and #6721: open PRs of an earlier attempt that add tests to 8 of these 13 cards. Their tests restate each constant (assert NAME == "agent-autopilot-L1118"), which this lane rules out; they are superseded, not built on.

Done when

• each card has these tests, passing under the reference (t27c test-report, 0 vacuous) and under t27b;
• each is sealed;
• the ledger rows that move come from bless of the branch run;
• the signed lane receipts compare as IMPROVED_ONLY.

Boundary

• specs/crons/README.md