test: cover pure-logic gaps found by the coverage audit

I ran make coverage and worked the report function by function, separating real gaps from interactive/IO wrappers that aren't unit-test targets. These tests fill the genuine pure-logic holes: predicates, parsers, formatters, transforms, and three modules that had no test file at all.

New files cover car-member (local-repository), show-kill-insert-item (show-kill-ring), the oauth2-auto plstore cache fix (auth-config), the coverage-core project-root fallback, reconcile--dirty-p, and the recurrence-frequency dispatch in calendar-sync. Extended files add the missing branches: coverage-core's merge-base and diff /dev/null handling plus the staged and branch-vs-main scopes, the detect-system-timezone symlink path, user-constants no-op and optional-failure branches, the elfeed playlist branch with HTML-entity decoding, the duplicate-line no-comment-syntax guard, and several calendar-sync edges (exception field overrides, timestamp seconds and TZID fallback, property-line position advancement, parse-ics nil and out-of-range inputs).

Mocks sit at the real boundaries (plstore, url-retrieve, process-file, git) so each function's own logic runs. Dates come from relative helpers. About 65 tests added across 15 files, and the full suite stays green.

0 files changed, 0 insertions, 0 deletions