feat(pm): dispatch-gates names the CI workflow jobs a card's paths schedule (#17094)

* feat(pm): dispatch-gates names the CI workflow jobs a card's paths schedule

The derivation's vocabulary was `check:*` families, and a workflow job
watches a path just as much as a gate script does. Of the 74 workspace
packages that declare a `test` script, zero are run by any of this tree's
215 `check:*` scripts -- every one is run by `Test Core (N/6)` or
`Dogfood Regression Gate (N/3)`. So a change set under such a package
could derive a complete family list, run all of it, reconcile it with
`--ran`, and truthfully report full coverage while a test in the very
package it edited was red in CI.

`jobFilteredSteps` reads the job populations this tool already resolves
for the family derivation and reports, per job your paths schedule, the
steps no discovered family accounts for. It is the complement of the
always-runs tail: that one reads the jobs CI cannot narrow by path, this
one reads the jobs your paths narrow it to.

The block sits outside the runnable total and never reaches `--commands`
stdout -- every row is CI's own shell in CI's environment. It is named on
all three lanes through `outsideBlockNames`, the one place the list of
outside blocks exists.

Claude-Session: https://claude.ai/code/session_01HxLw5aKDPR5RJgyUR7Exkd
Co-authored-by: Claude <noreply@anthropic.com>

* test(pm): the new self-test cases degrade to failures instead of aborting the battery

The ablation that proves these pins can fail removes the very fixture row
they read, and a TypeError there stops the battery before the LIVE
controls underneath it are decided -- turning "this pin can fail" into
"this pin was never reached".

Claude-Session: https://claude.ai/code/session_01HxLw5aKDPR5RJgyUR7Exkd
Co-authored-by: Claude <noreply@anthropic.com>

---------

Co-authored-by: Claude <noreply@anthropic.com>

Author: yinlianghui