feat: start healthy and retry when Plex is unreachable at boot

[cplieger]

Summary

Makes plex-language-sync survive a Plex that is down at boot instead of crash-looping, and enrolls it on the shared image-smoke harness.

Previously the initial Plex connect + admin resolution exited non-zero on any failure, so under restart: unless-stopped a Plex outage at startup crash-looped the container. Startup now classifies the failure (mirroring plex-exporter):

• Fatal (bad PLEX_TOKEN/4xx, wrong-server 404, TLS/cert error) -> exit non-zero; the container never goes healthy, so the misconfiguration stays loud.
• Transient (Plex unreachable, DNS/timeout, or 5xx) -> mark the container healthy-degraded and retry the initial connect with capped exponential backoff (1s->30s) until Plex answers, then resolve the admin and begin normal operation.

WebSocket disconnects after the initial connect are unchanged (the listener already reconnects with backoff).

Changes

• internal/plex/client.go: add a typed HTTPStatusError (carries the status code; message text unchanged) so the classifier can distinguish 4xx from 5xx.
• main.go: connectAndResolveAdmin retry-until-connected loop + isFatalStartupError classifier + capped backoff; DeriveKey moved ahead of the connect so a bad token is surfaced before any healthy-degraded marking.
• main_test.go: table test for the classifier + backoff.
• README health section updated for the new semantics.

Smoke test

Enrolls plex-language-sync on the shared configs/image-smoke.sh harness (tests/image-smoke.conf, env-only dummy PLEX_URL/token). Validated locally: the image builds and the harness reports healthy after 5s against an unreachable Plex. Removes plex-language-sync's documented smoke-test exception.

Verification

go build, go vet, go test -race ./..., and golangci-lint run / fmt --diff all clean.