Cache the keycache location resolution instead of redoing it per operation

[djw8605]

Problem

resolve_cache_location() runs for every keycache operation — at least once per token validation, twice for validations that also store keys. Each call performs:

1. getpwuid_r (password-database lookup),
2. two mkdir calls,
3. a full SQLite open / CREATE TABLE IF NOT EXISTS / close round trip (initialize_cachedb),

…after which the caller opens the database a second time to do the actual query. That's two SQLite opens plus a DDL statement on the hot path of every validation, even with a warm cache.

Fix

• Split the existing logic into resolve_cache_location_uncached() and a caching wrapper.
• The cache is invalidated via a generation counter that configuration changes affecting the location (keycache.cache_home, keycache.allow_in_memory) bump.
• Failed resolutions are not cached, so transient permission/disk problems keep being retried.

Also sets the SQLite busy timeout in get_all_issuers_from_db() — the only DB path missing it — so a concurrent write no longer makes the background refresher silently return a partial issuer list.

Behavior note

A change to $XDG_CACHE_HOME after the first resolution is no longer picked up automatically (setting keycache.cache_home still invalidates immediately). Mutating the environment mid-process was of dubious reliability before.

Testing

• ctest unit, env_config, and monitoring suites pass — including SetGetConfiguredCacheHome/GetKeycacheLocation, which change the cache home mid-run and exercise the invalidation path.

🤖 Generated with Claude Code