fix(ollama): wrap timeout/connection errors so they retry

[dev-ploy]

Summary

• Fix page_generation_failed floods with local LLMs (Ollama): APITimeoutError
  and APIConnectionError are siblings of APIStatusError under
  openai.APIError, so the old catch-only-APIStatusError clause let them escape
  uncaught, skipping the tenacity retry loop entirely and surfacing as
  page_generation_failed in orchestrate.py (run_level → guarded_named).
• Move the Ollama provider onto the shared retry policy
  (provider_should_retry / provider_retry_stop / provider_retry_wait) with a
  3-way exception mapping (RateLimitError → 429, APIStatusError → HTTP status,
  APIError → transient) so transient timeout/connection errors retry while
  non-retryable 4xx (400/401/403/404/405/413/422) fail fast. Same wrapping applied
  to both stream_chat sites.
• Apply the same defensive APIError catch to the other OpenAI-client providers
  (openai, openrouter, deepseek), and warn when a local provider
  (ollama / codex_cli / opencode) runs with --concurrency > 4.

Related Issues

Closes #445

Test Plan

• Tests pass (pytest) — 897 provider + CLI tests green
• Lint passes (ruff check) on all changed files
• Web build passes (npm run build) — N/A, no frontend changes

Additional manual verification against a live Ollama server, across multiple
locally running models:

• Full repowise init --provider ollama --concurrency 6 --coverage 0.5 runs
  against several local models → all pages generated with 0
  page_generation_failed, and the local-provider concurrency warning shown.
• Forced a real APITimeoutError (1 ms client timeout against the live server):
  now wrapped as ProviderError and retried 3× instead of escaping raw.
• New regression tests confirm 401 (auth) and 400 (invalid request) fail on the
  first attempt — no wasted retries.

Checklist

• My code follows the project's code style
• I have added tests for new functionality
• All existing tests still pass
• I have updated documentation if needed

[repowise-bot]

✅ Health: 7.6 (unchanged)

📋 At a glance
1 file changed health · 5 hotspots touched · 6 new findings introduced · 5 co-change pairs left out · 3 files with recent fix history. Scoped to packages.

⚠️ Change risk: moderate _{(riskier than 52% of this repo's commits · raw 9.0/10)}
This change's risk is driven by:

• more lines added than baseline
• more scattered than baseline

🩹 Review priority _{(files here with the most recent bug-fix history — defects cluster, so review these first)}

• .../init_cmd/command.py — fixed 3× in the last ~6 months
• .../llm/openai.py — fixed 2× in the last ~6 months
• .../llm/deepseek.py — fixed once in the last ~6 months

File	Score	Δ	Why
.../init_cmd/command.py	2.7 → 2.8	▲ +0.2	✅ resolved function hotspot, complex method, dry violation

🔎 More signals (2)

🔥 Hotspots touched (5)

• .../init_cmd/command.py — 19 commits/90d, 1 dependents · primary owner: Raghav Chamadiya (93%)
• .../llm/deepseek.py — 6 commits/90d, 3 dependents · primary owner: ecourn (59%)
• .../llm/ollama.py — 4 commits/90d, 5 dependents · primary owner: Raghav Chamadiya (93%)

2 more

• .../llm/openrouter.py — 5 commits/90d, 8 dependents · primary owner: lenya lozhkin (55%)
• .../llm/openai.py — 6 commits/90d, 8 dependents · primary owner: Raghav Chamadiya (87%)

🔗 Hidden coupling (2 files)

• .../init_cmd/command.py co-changes with these files (not in this PR):
  • docs/CLI_REFERENCE.md (9× — 🟢 routine)
  • .../init_cmd/workspace.py (7× — 🟢 routine)
  • docs/USER_GUIDE.md (5× — 🟢 routine)
  • .../cli/helpers.py (5× — 🟢 routine)
• .../llm/openai.py co-changes with .../llm/anthropic.py (6× — 🟢 routine) — not in this PR.

👀 Suggested reviewers @ecourn

---

📊 Full report · ⭐ Star Repowise · 📥 Install bot · Last updated 2026-07-14 16:01 UTC
_{Silence on a single PR with [skip repowise] in the title · Per-repo toggle on repowise.dev/settings?tab=bot}

[dev-ploy]

Hi @RaghavChamadiya i would love your view on this solution

[RaghavChamadiya]

Really solid fix, thanks, and it held up well despite sitting for a bit. I confirmed the bug is still live on main: Ollama still catches only APIStatusError, so APITimeoutError and APIConnectionError keep escaping tenacity and surfacing as page_generation_failed. The root-cause framing is spot on, and moving Ollama onto the shared provider_should_retry / provider_retry_stop / provider_retry_wait policy is exactly right, it lines up with what the openai provider already does.

I pulled the branch: it still merges cleanly, the full provider suite is green (166 passing), and the new regression tests pin the important parts, both that timeout/connection errors now retry and that non-retryable 4xx still fail fast. Appreciate the manual verification against a live Ollama server too.

Merging.