fix(pr): count failed merges and disjoin them into the exit code - #907

What this fixes

gr pr merge reported success when every merge in a run had failed. The per-repo diagnostics were printed and truthful; the exit code was not. A script driving the command was told a batch had merged when none of it had.

Three exits returned Ok while carrying failures:

Each now fails with a count of what failed against what was attempted.

"We looked and found nothing" and "we could not look" are different answers, and only the first is an absence of PRs. Lookup failures are now tracked separately from skips rather than arriving at the summary as one state.

The fixtures had to be corrected first, and that is the interesting part

mock_get_pr mounts a single invariant response. A read issued after a successful merge still reported merged: false — a state the real API cannot produce.

Three groups of tests were affected here, by three different causes. They have overlapping membership and nearly equal size, so counting them together is easy and wrong; each is stated separately below.

Five tests: failing because of the fixture, fixed with no assertion touched

Five tests began failing the moment the exit code became truthful, and they were failing because of the fixture rather than because of the change.

mock_pr_lifecycle couples the GET and the merge PUT through shared state so the sequence GET → PUT → GET behaves as the live API does. All five then passed with no assertion changed, which is what distinguishes a fixture defect from a contract change. Had those assertions needed editing, that would have been evidence the exit-code change was breaking real contracts rather than fixing a lie.

One test: the same fixture defect, never red, and it gains an assertion

test_pr_merge_repo_filter_excludes_non_target carried the same defect in the opposite direction and never failed at all. Its PR was mocked as state "open" and merged: true simultaneously, which made the command's own post-merge verification vacuous — it passed whether or not the merge did anything.

It now starts unmerged, and it gains an assertion that the merge actually fired. So it is not one of the five: its fixture changed, it was never red, and its assertion count went up rather than staying fixed.

One test: a genuine contract change, and the only assertion this PR edits

test_pr_merge_branch_behind_suggests_update was red for an unrelated reason and its fixture is untouched. It required is_ok() while its own failure message read "handled without crashing". Those are different claims. The merge did not happen, so graceful handling means a useful error rather than a success.

That is the whole distinction this PR turns on: five fixture-caused failures fixed with zero assertion edits, and exactly one genuine contract change fixed by editing exactly one assertion.

Mutation evidence

Neutering each of the three exits independently turns a named test red:

The first mutation pass found three of the guards unwitnessed. The --auto path had no test of any kind, so its guard arrived together with the first test that path has ever had.

Verification

• Full suite: cargo test exit 0, 56 result lines, zero failing binaries.
• cargo fmt --all --check: differential against a base checkout of the target. Base flags one pre-existing unrelated file; head adds nothing. Only the three changed files were reformatted, deliberately not the repository, so no unrelated file enters this range.

Not in scope

The "No open PRs found for any repository." wording is owned by another change in flight and is untouched here. The pre-existing --wait timeout exit is also unwitnessed; it predates this change and is left alone rather than fixed silently.

Premium boundary: gitgrip is OSS. This is workspace orchestration and CLI exit-code behavior, with no identity, org, or entitlement semantics.

Ref #884 — closes at promotion
Ref #886 — closes at promotion