feat(algorithm): expose owned training status

[minsing-jin]

Closes #4

Summary

• Adds a truthful owned-algorithm training status surface for Hedwig: default algorithm.yaml priors, local logistic SGD weights, and optional LightGBM LambdaMART readiness.
• Reports dependency/model availability, feedback/history sufficiency, active LTR backend, enabled components, optional SOTA flags, and recent retrain/adoption history on /status.
• Clarifies README and Hybrid Ensemble docs that SOTA backends are optional and dependency/model-gated, not required for local use.
• Keeps fallback behavior non-fragile: missing LightGBM, missing model files, or sparse feedback fall back to local logistic/default-prior ranking with explicit user-visible status.
• Keeps Manus/API delegation and LiteLLM out of scope.

Verification

• git diff --check
• python3 -m pytest tests/test_algorithm_training_status.py -q -> 4 passed
• python3 -m pytest tests/test_algorithm_training_status.py tests/test_v3_meta_retrain.py tests/test_v3_phase8_and_phase7_full.py -q -> 30 passed
• python3 -m pytest -q -> 722 passed, 1 warning
• Ouroboros QA qa-9942a847 -> 0.93 PASS

Notes

• PR branch was rebased onto the merged feed-moat stack from PR feat(feed): add personal algorithm policy engine #18.
• The earlier broad v3 commit was removed from this PR so feat(algorithm): expose owned training status #5 is scoped to issue feat(algorithm): harden owned learning engine and SOTA training path #4 only.
• The first full test run in the /private/tmp worktree failed only because that worktree had no .venv/bin/python; after linking the existing project .venv, targeted CLI tests and the full suite passed.