v0.10.1

This release makes i2scim's Signals client behave the way operators of goSignals expect, and ensures pending events and acks survive a restart.

• Streams now report a clear lifecycle state

  • Streams are enabled, paused, or disabled instead of a single error flag. paused self-recovers; disabled means an operator needs to look. Existing ssfConfig.json files upgrade automatically.
  • HTTP failures are handled by code: 401 retries briefly, 403 stops immediately, 429 honors Retry-After and never auto-disables, 5xx and network errors back off and retry.
  • On push or poll failure, i2scim asks the remote's /status endpoint whether it has paused itself — distinguishing a remote outage from a remote that's deliberately quiet.
  • A keepalive event is sent on idle push streams, so a broken path is caught before real events pile up.
  • Protocol errors from the receiver (RFC 8935 §2.4) are recorded with the specific error code and JTI in errorMsg. A signature failure automatically reloads the issuer PEM and retries once.
  • New config: scim.signals.pub.unauthorized.retry.{max,delay}, scim.signals.pub.status.check.interval, scim.signals.pub.idle.verify.interval, plus rcv.* equivalents.

• Pending events and acks survive restart

  • Failed push events are persisted — to MongoDB on the Mongo backend, to <scim.prov.memory.dir>/events/ on the memory backend. Restarts no longer lose them, and a disabled stream keeps its queue for automatic replay when re-enabled.
  • SCIM operations no longer wait on push outcome. A slow receiver can no longer slow down the SCIM API.
  • Retry budgets are now wall-clock based (default 6 hours) instead of attempt counts.
  • Poll-side acks are persisted too, so a restart between "applied locally" and "ack delivered" doesn't cause re-delivery.
  • The issuer PEM file is watched for changes — out-of-band key rotations are picked up automatically.
  • Events generated before stream registration completes are buffered and replayed once registration succeeds. A persistent failure raises clear warnings.
  • Disk and queue usage is monitored with configurable warn / critical / fatal thresholds.
  • New config: scim.signals.pub.retry.elapsed.limit, scim.signals.pub.pem.watch, scim.signals.pub.storage.{warn,crit,fatal}.pct, scim.signals.pub.preregister.warn.minutes.

• Security and fixes

  • Patched CVE-2026-39852 and CVE-2026-41417; refreshed dependencies.
  • JWKS load failures at boot are now non-fatal — a temporarily unreachable JWKS no longer blocks startup.
  • /status probes append the stream_id query parameter per SSF §7.1.2.
  • Fixed NPE on receipt of inbound verify events.
  • Quieted noisy CycloneDX and retry-worker shutdown warnings.