perf(temporal): reuse the prediction HTTP client

[frgfm]

Reuse one HTTP client per API worker. Close it after the validation worker stops.

flowchart LR
    A[Create once] --> B[Reuse for predictions] --> C[Close at shutdown]

Loading

Warm median per prediction	main	This PR	Less time
HTTP	9.93 ms	1.47 ms	85%
HTTP, server closes every connection	8.32 ms	2.11 ms	75%
HTTP, simulated 50 ms inference	60.29 ms	51.76 ms	14%

Actual predict(), local TCP, Python 3.11.15/httpx 0.28.1; median of 5 fresh-process runs × 100 calls after warmup. Inference is simulated; first-call speedup is unproven.

Benchmark sources and raw results.

Scope: +10 net production lines (+29 total), one new test. Independent of #705, #706 and #707; merge checks pass.

Checks: 25 focused tests, real TCP recovery/shutdown probes, lint, typing, pre-commit and full CI pass.

[codecov]

Codecov Report

✅ All modified and coverable lines are covered by tests.
✅ Project coverage is 93.89%. Comparing base (25f3d2e) to head (48eef0e).

Additional details and impacted files

@@            Coverage Diff             @@
##             main     #708      +/-   ##
==========================================
+ Coverage   93.87%   93.89%   +0.01%     
==========================================
  Files          59       59              
  Lines        3218     3226       +8     
==========================================
+ Hits         3021     3029       +8     
  Misses        197      197              

Flag	Coverage Δ
backend	94.00% <100.00%> (+0.01%)	⬆️
client	91.30% <ø> (ø)

Flags with carried forward coverage won't be shown. Click here to find out more.

☔ 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.