feat: add InferenceHub provider to OpenAI-compatible registry

[tianyiswufeng]

What

Adds inferencehub to the OpenAI-compatible provider registry:

• aimux-providers/src/provider_registry.json — new entry (sorted position):
  name: inferencehub, display: InferenceHub, base_url: https://app.inferencehub.tech/v1, env_var: INFERENCEHUB_API_KEY, profile: {}
• Every other changed file is regenerated output: scripts/gen_provider_names.py (ProviderName types for Rust + the 7 binding languages) and scripts/gen_providers_doc.py (docs/api/providers.md).
• Bumped the two hardcoded registry-count asserts in provider.rs tests (registry_entries_are_valid, provider_name_roundtrip) 250 → 251.

Note: README says "325 providers / 250 registry-backed", while the generated docs/api/providers.md now counts 327 / 251 (it was 326 / 250 before this PR, so the total was already off by one). I left README untouched rather than half-fix it — happy to update those numbers here too if you'd like.

Protocol evidence (per RFC-0006)

• Official API docs: https://inferencehub.tech/docs/api — documents POST /v1/chat/completions and GET /v1/models on base https://app.inferencehub.tech/v1, Authorization: Bearer auth, and SSE streaming ("stream": true).
• Shared-layer reuse basis: standard OpenAI wire; no deviations observed, so the default profile ({}) applies and no shared-layer changes are needed.

Verified against the live endpoint

Ran the exact registry code path via the Node binding's provider(...) with a baseUrl override before adding the entry:

• generateText ✓ (text + usage) — model zai-glm-5.2
• streamText ✓ (incremental deltas) — model deepseek-v4-flash
• tool calling ✓ (tool_calls returned, finish reason tool-calls)
• GET /v1/models returns the live model id list (e.g. zai-glm-5.2, deepseek-v4-flash, gpt-5.6-sol)

Tests

No-difference thin wrapper → no cassette added (RFC-0006: "A no-difference thin wrapper does not need to record a cassette merely for process completeness"). Locally: cargo fmt --all -- --check, cargo clippy -p aimux-providers --all-targets -- -D warnings, and cargo test -p aimux-providers --tests pass.

Disclosure: I run InferenceHub. Happy to adjust anything about the entry.

🤖 Generated with Claude Code

[eric8810]

could you please share some knowledge into our provider inventory documentation

[tianyiswufeng]

Added in 6ff3f7a — provider-inventory/providers.json + providers.csv now carry an inferencehub entry, and it's appended to aimux_provider_modules.

Notes on the field choices, so you can adjust anything to taste:

• Shaped after the registry-backed aggregator entries (aihubmix-class): provider_kind: gateway_aggregator, tier L2 / protocol: openai / openai_compatible: true, implemented_in_aimux: true.
• Set trust.status: "review" rather than self-assigning accepted — per the inventory README, new entries need your manual endpoint/protocol/legal review. Provenance is recorded with source_id: "provider_submission" pointing at this PR, with the docs URL, INFERENCEHUB_API_KEY, and endpoint metadata filled in (models_url is the real documented endpoint, so models_url_inferred: false; note GET /v1/models requires auth).
• Capabilities chat, embedding, image_generation, rerank and the model count/sample (43 ids) match the official docs (https://inferencehub.tech/docs/api) and the live /v1/models as of Aug 3. There's also an Anthropic-compatible /v1/messages surface on the bare origin (https://app.inferencehub.tech) that works with your native anthropic provider via base_url / ANTHROPIC_BASE_URL — I kept the inventory entry OpenAI-shaped since that's what the registry entry uses; happy to record it as mixed-protocol instead if you prefer.
• Left summary counts, the inventory README totals, and raw-provider-records.jsonl untouched — those look pipeline/generator-owned (codex and azure_ai_foundry aren't in the raw records either). Say the word if you'd like any of them bumped here.

[eric8810]

will merge when ci is complete