fix(daemon): recover stash target oid when last stash is popped

[posthog]

Problem

• Users who pop or drop their last remaining stash trip two error-level daemon exceptions every time — grouped by message, ~150 across 4 users in a month, on Linux, macOS, and Windows.
• Git deletes both .git/refs/stash and .git/logs/refs/stash when the last stash leaves the stack. Both trace-payload resolvers read those files from disk, so both fail on the same command and each logs at error level; the telemetry bridge turns each into an $exception. That is why the two issues always arrive as a pair, ~1s apart, for the same worktree.
• Per-issue counts look tiny only because the worktree path enters the fingerprint, which splits one bug across many issues.

Changes

• Route both payload resolvers through a shared recover_top_stash_target_oid_from_history helper that reuses the daemon's existing rewrite-history fallback (inferred_top_stash_sha_from_rewrite_history) — the same recovery the event builder already relied on.
• resolve_stash_target_oid_for_command and the pop/drop/branch arm of resolve_stash_target_oid_for_terminal_payload: on an on-disk miss for a top-of-stack target, recover the sha from history, or return a benign None (which the event builder still fills) instead of erroring.
• Non-top-of-stack targets (e.g. stash@{2}) still return an error — those genuinely cannot be recovered from history.
• Point the pre-existing StashOperation::Apply fallback at the new helper too, so there is one copy of the recovery expression.

Impact

• No behaviour change to authorship attribution — the event builder already had the fallback, so the final data was already correct. This removes the error-level telemetry noise and populates the intermediate autter_stash_target_oid on the affected traces.

Tests

• Added two lib unit tests: one asserts both payload resolvers recover the target sha when refs/stash and its reflog are absent; one asserts a non-top-of-stack reference still errors.
• cargo check --locked --all-targets (the CI job) and the existing stash integration suite pass.

Agent context

• Considered folding the top-of-stack gate + error arms into one shared function, but the two resolvers carry distinct error messages, so a recover_* helper plus an inline gate reads cleaner than a combinator.
• Left the rare unrecoverable-and-terminal double read of the rewrite log alone: the common recoverable path reads once (the first resolver succeeds and the second is skipped), and memoizing across the large augmentation function would add cross-cutting state beyond this fix.

---

Created with PostHog Desktop from this inbox report.

---

^{Need help on this PR? Tag @codesmith-bot with what you need. Autofix is disabled.}