Feature/decouple websocket

[sambles]

Needs: OasisLMF/OasisLMF#2105

Fixes to run platform without redis websocket

New: HTTP progress endpoint

• src/server/oasisapi/queues/views.py (new) — AnalysisStatusView, an unauthenticated
  APIView at /analysis-status/ (outside the versioned v1/v2 API, alongside
  healthcheck//server_info/). Applies events_complete as an atomic F() SQL increment so
  concurrent worker processes can't clobber each other's updates.
• src/server/oasisapi/queues/serializers.py — AnalysisStatusSerializer (analysis_pk,
  optional events_total/events_complete).
• src/server/oasisapi/base_urls.py — wires /analysis-status/ in at the root, alongside
  healthcheck//server_info/ (moved out of the versioned v2 router it started in).
• src/server/oasisapi/queues/consumers.py — send_task_status_message now short-circuits
  when settings.DISABLE_WORKER_WS is set, skipping the websocket broadcast entirely.

Hardening: unreachable channel layer no longer crashes callers

send_task_status_message's DISABLE_WORKER_WS check only helps when someone remembers to set
it — e.g. the debug compose file drops the channel-layer (redis) service but that flag needs
to be set separately, and nothing enforces the two changes travel together. Without it, an
unreachable channel layer raised straight out of group_send(...), uncaught by any of this
function's callers — some of which (AnalysisTaskStatusQuerySet.create_statuses() in
analyses/models.py, called when a run's chunk sub-task rows are created) sit on the actual run
lifecycle, not just UI notification. That could fail or error out a real analysis purely because
of a broken websocket broadcast.

• src/server/oasisapi/queues/consumers.py — send_task_status_message now wraps the
  group_send(...) call in try/except redis.exceptions.RedisError, logging a warning and
  returning instead of propagating. This complements DISABLE_WORKER_WS (an explicit "don't even
  try" opt-out) by also covering redis being down, misconfigured, or unreachable for any other
  reason — no reliance on remembering to set the flag correctly.

Bug fixs: progress reset race across chunks

events_total is reported once per chunk, not once per analysis (each chunk is its own
worker process running generate-losses-chunk). The original code reset num_events_complete
to 0 every time an events_total ping arrived, so a later-starting chunk would wipe out
progress already reported by chunks further along in the same run.

Fixed by moving the reset to the one place that unambiguously means "a run (or re-run) is
starting":

• src/server/oasisapi/analyses/models.py — Analysis.run() now resets
  num_events_total = 0 and num_events_complete = 0 at the same point it sets
  status = RUN_QUEUED.
• src/server/oasisapi/queues/views.py / consumers.py — events_total pings now
  only update num_events_total; they no longer touch num_events_complete.

[codecov]

Codecov Report

❌ Patch coverage is 60.00000% with 16 lines in your changes missing coverage. Please review.
✅ Project coverage is 74.06%. Comparing base (e46412b) to head (8db6abe).
⚠️ Report is 1 commits behind head on main.

Files with missing lines	Patch %	Lines
src/server/oasisapi/queues/views.py	47.82%	12 Missing ⚠️
src/server/oasisapi/queues/consumers.py	55.55%	4 Missing ⚠️

Additional details and impacted files

@@            Coverage Diff             @@
##             main    #1427      +/-   ##
==========================================
- Coverage   74.10%   74.06%   -0.04%     
==========================================
  Files         214      215       +1     
  Lines       14140    14178      +38     
==========================================
+ Hits        10478    10501      +23     
- Misses       3662     3677      +15     

☔ View full report in Codecov by Harness.
📢 Have feedback on the report? Share it here.

🚀 New features to boost your workflow:

• ❄️ Test Analytics: Detect flaky tests, report on failures, and find test suite problems.

[Ha-Ree]

Should add another

self.num_events_total = 0
self.num_events_complete = 0
self.save()

[Ha-Ree]

Claude:
DISABLE_WORKER_WS=true silently disables k8s autoscaling. send_task_status_message feeds both the UI and the worker-controller autoscaler, which subscribes to ws/v2/queue-status/ (kubernetes/worker-controller/src/oasis_websocket.py L34). The flag is safe in compose but in Helm it would stop scaling with only a DEBUG log. Needs a loud warning at startup, or the flag scoped to the analysis-status broadcast only.