v0.5.0

[0.5.0] - 2026-09-05

Added

• The trial length is the server's, not the build's. A tenant could set a
  trial length in the dashboard and nothing happened to their Rust app:
  KeylightConfig::trial_duration_days and free_tier_enabled were read
  directly wherever a trial decision was made, so the value compiled into the
  build was the only one that ever applied. New
  Keylight::effective_trial_duration_days() and
  Keylight::effective_free_tier_enabled() resolve server value → local seed
  → 0, and check_trial() / state() now read those. The config fields stay
  as a seed, deliberately: a brand-new install genuinely has nothing else, and
  removing them would make first-launch behaviour depend on the network.
• The settings ride on calls already being made, so launch-time network I/O
  stays at zero. validate covers every licensed install and the keyless beacon
  covers every unlicensed one; both responses now carry trial_duration_days
  and free_tier_enabled, and both are absorbed. Keylight::fetch_config()
  reads the dedicated GET /{tenant}/{product}/config route for hosts that want
  an explicit refresh — it is not on the launch path, and a test asserts
  that resolving state never touches it.
• sdk_trial_duration_days on activate and validate — the length the build
  was compiled with, not the effective one, since echoing the server's own
  number back diagnoses nothing. It catches the ordinary mistake of a 30-day
  build against a 14-day dashboard setting. Diagnostic only: the server must
  never gate on it, because a patched client sends whatever its author wants.

Fixed

• 0 is a real setting and absence is not zero. The cached pair is stored
  as Options and merged field by field, so a worker that sends neither field
  leaves what the install already learned alone, a server 0 survives a
  relaunch as 0 rather than falling back to the seed, and a server
  free_tier_enabled: false is not mistaken for "never heard".
• start_trial() stamps the clock even at a zero duration. Once the
  duration is server-owned, 0 is indistinguishable from "the config has not
  arrived yet", so bailing out left no start timestamp for a later-arriving
  duration to measure and the user never got the trial their tenant enabled. The
  stamp grants nothing on its own — check_trial() still reports no trial while
  the effective duration is 0. An existing stamp is never overwritten, so
  enabling a trial 60 days after an install does not hand it a fresh window.

Notes

• No breaking change. KeylightConfig gains no field, so construction — builder
  or struct literal — is untouched. #[non_exhaustive] was deliberately not
  added: it would break the documented KeylightConfig { .. } literal that
  max_offline_days tells users to reach for, and the guard only earns its cost
  when a field is actually being added.
• 100 → 115 tests.
• Ports the contract shipped in keylight-cpp 0.2.0/0.2.1; see that repo's
  docs/superpowers/specs/2026-09-05-trial-parity-handoff.md.