feat(issue-42): add provider-neutral multi-subscription routing

[konard]

What this PR does

Extends multi-account routing from Claude-only pools to every subscription provider already supported by the router: Claude, Codex, Gemini, and Qwen. It also adds stable session assignment, strict token account pins, internal request caps, quota-aware selection/cooldowns, and Formal AI-style native protocol namespaces.

Root cause and reproduction

PR #38 added vendor credential readers and single-subscription forwarding, but AccountRouter still constructed Claude-only OAuthProvider instances. Codex, Gemini, and Qwen therefore bypassed account selection; refreshed credentials were cached only by provider; and OpenAI translation discarded session metadata before selection.

The regression tests added first reproduced those gaps as 26 missing API/type errors, including provider-neutral pool construction, strict session/pin selection, per-account refresh keys, request caps, and request-routing context extraction.

Routing behavior

• One provider-neutral pool backed by SubscriptionReader for Claude, Codex, Gemini, or Qwen.
• round-robin, priority / fill-first, and normalized least-used / quota-first selection for new sessions.
• Atomic ACCOUNT_REQUEST_LIMITS accounting, including concurrent cap enforcement.
• Session affinity copied from stable headers/body metadata before translation and renewed while active.
• Router-token account claims and existing session bindings are strict: unavailable identities fail instead of silently moving.
• Only HTTP 429 starts cooldown; standard Retry-After delta/date values extend the configured minimum.
• Refresh cache isolation by (provider, account) and token-budget enforcement on subscription routes.
• Account limits and remaining capacity in /v1/accounts, accounts list, and doctor.

Native protocol routes

Adds namespaced Anthropic, OpenAI, Codex, and Qwen aliases plus native Gemini generateContent / streamGenerateContent and Vertex publisher-model paths. Existing unprefixed APIs remain unchanged. Provider-specific namespaces use the matching configured UPSTREAM_PROVIDER; no cross-provider billing or protocol fallback is introduced.

Research and requirement trace

The case-study package at docs/case-studies/issue-42/ records immutable Claudexor and Formal AI snapshots, related-project research, adopted design rules, deliberate boundaries, and an R1-R17 implementation/test trace.

Testing

• cargo fmt --all -- --check
• cargo clippy --all-targets --all-features -- -D warnings
• cargo test --all-features — 137 unit, 50 integration, and 10 release-workflow tests pass; doc tests pass
• cargo doc --no-deps --all-features
• repository-wide Rust file-size policy — every file is under 1,000 lines
• real cargo run -- serve startup — Axum route table starts without conflicts
• CLI smoke checks for accounts list and doctor

This is a backend/configuration change, so no UI screenshots apply. The minor-release changelog fragment is included; release automation owns the version bump.

Closes #42

[konard]

Working session summary

Completed and marked ready for review: PR #43.

• Added provider-neutral Claude, Codex, Gemini, and Qwen subscription pools with routing strategies, session affinity, request caps, quota tracking, and cooldowns.
• Added native provider namespaces and Gemini/Vertex routes.
• Added regression tests and an issue case study, informed by pinned Claudexor and Formal AI snapshots.
• Local formatting, Clippy, documentation, runtime smoke checks, and all tests passed.
• Final-SHA CI passed on Ubuntu, macOS, and Windows.
• Branch is synchronized with main, pushed, clean, and mergeable.

---

This summary was automatically extracted from the AI working session output.

[konard]

🤖 Solution Draft Log

This log file contains the complete execution trace of the AI solution draft process.

💰 Cost estimation:

• Model: GPT-5.6 Sol
• Provider: OpenAI
• Public pricing estimate: $33.058331

📊 Context and tokens usage:

• 773.7K / 200K (387%) input tokens, 88.7K / 128K (69%) output tokens

Total: (773.7K + 21.3M cached) input tokens, 88.7K output tokens, $33.058331 cost

🤖 Models used:

• Tool: OpenAI Codex
• Requested: gpt-5.6-sol
• Thinking level: xhigh (~31999 tokens)
• Model: GPT-5.6 Sol (gpt-5.6-sol)

📎 Log file uploaded as Gist (10575KB)

• View complete solution draft log

---

Now working session is ended, feel free to review and add any feedback on the solution draft.

[konard]

✅ Ready to merge

This pull request is now ready to be merged:

• All CI checks have passed
• No merge conflicts
• No pending changes

---

Monitored by hive-mind with --auto-restart-until-mergeable flag