ci.yml PostgreSQL setup retry loops: SC2015/SC2034 shellcheck findings (test job, lines 55-56, 64)

[cdeust]

What

actionlint (which also runs shellcheck against embedded run: blocks)
flags the test job's "Set up PostgreSQL + pgvector (runner-local, no
registry pull)" step (.github/workflows/ci.yml:44-65):

$ actionlint .github/workflows/ci.yml
.github/workflows/ci.yml:45:9: shellcheck reported issue in this script: SC2015:info:10:40: Note that A && B || C is not if-then-else. C may run when A is true [shellcheck]
.github/workflows/ci.yml:45:9: shellcheck reported issue in this script: SC2015:info:11:76: Note that A && B || C is not if-then-else. C may run when A is true [shellcheck]
.github/workflows/ci.yml:45:9: shellcheck reported issue in this script: SC2015:info:20:65: Note that A && B || C is not if-then-else. C may run when A is true [shellcheck]
.github/workflows/ci.yml:45:9: shellcheck reported issue in this script: SC2034:warning:19:1: i appears unused. Verify use (or export if used externally) [shellcheck]

Three for i in 1 2 3; do <cmd> && break || sleep 5; done / for i in $(seq 1 30); do <cmd> && break || sleep 1; done retry loops
(lines 55, 56, 64) use A && B || C, which is not if/then/else: if B
itself fails (e.g. break inside a loop that a shell option makes fail, or
any future edit that puts a failing command in the B slot), C runs too.
Today break cannot itself fail, so this is not a live bug, but it is exactly
the shape SC2015 exists to flag before an edit turns it into one. $i is
also unused beyond the loop counter (SC2034), which is usually a false
positive for a retry counter but is worth silencing explicitly rather than
leaving a standing warning.

Why filed here, not fixed inline

Found 2026-07-29 while verifying an unrelated .github/workflows/ci.yml edit
for issue #251 (a new step in the lint job, far below this one in the
test job). actionlint/shellcheck is not part of this repo's own gates
(grep confirms zero references in .github/, no pre-commit config) — I ran
it as an extra precaution on my own new YAML, and it surfaced this
pre-existing, unrelated finding. None of #251's own build/lint/test steps
reach this code path, so it is outside that change's blast radius per
coding-standards.md §14.3, and is deferred here rather than folded into an
unrelated PR (which would also add merge-conflict surface across the several
other worktree branches presently editing ci.yml).

Acceptance criteria

• Each A && B || C retry loop (lines 55, 56, 64) rewritten as an explicit
  if/until so a failing B cannot fall through to C.
• The retry-loop counter variables are either used (e.g. logged on final
  failure) or renamed/prefixed to signal intentional non-use, silencing
  SC2034.
• actionlint .github/workflows/ci.yml reports zero findings.
• The test job still passes on Python 3.10-3.13 (no behavior change to the
  PostgreSQL provisioning retry semantics).

[cdeust]

Resolved by #271 (merged as b10b43a), which fixed every actionlint/shellcheck finding in ci.yml and wired actionlint into the Lint job.

Verified on main @ 56f2f4f: actionlint .github/workflows/*.yml reports 0 findings, and 0 occurrences of SC2015/SC2034 in ci.yml (this issue's subject). actionlint is referenced 11 times in ci.yml, so a reintroduction now fails CI rather than being caught by hand.

Noting for the record: this issue was filed by #268 as a deferral for lines that a sibling PR (#261/#271) was already fixing — the code-reviewer flagged the redundancy during review.