fix(ci): retry only the answers that can change on a second attempt

[yunze7373]

Problem

The previous change classified every 4xx as final. A 429 rate limit or a
408 timeout would then fail the release on the first answer, even though both
can succeed moments later.

An intermediate version tried to handle this with an empty 408 | 429) branch,
which does not work: ;; ends the branch instead of falling through, so a
retryable status would have left the while loop spinning with no delay and no
attempt counter.

Changes

• Retry 000 (transport failure), 408, 429, and 5xx; report anything else at once.
• Move the retry bookkeeping out of a case branch to after the case statement, so no status can skip the delay or the attempt count.

Verification

The retry loop was extracted from the parsed workflow with the HTTP call
replaced by a stub that returns a scripted sequence of status codes:

Scenario	Exit	Retried	Outcome
202	0	0	accepted
403	1	0	not retried
401	1	0	not retried
404	1	0	not retried
429 then 202	0	1	accepted
500 then 202	0	1	accepted
000 then 202	0	1	accepted
408 throughout	1	2	failed after 3 attempts
503 throughout	1	2	failed after 3 attempts

npm run lint clean, npm test 140/140.