[senamakel]

Summary

Adds TinyHumans managed inference support to TinyInference through the OpenAI-compatible provider: live bearer resolution, managed usage and reasoning projection, optional thread and attribution headers, readiness probing, and model catalog discovery. The adapter includes request model selection and native tool-calling controls and reports authentication failures through an operation-aware callback.

The follow-up also fixes the initial hosted CI compile errors by explicitly selecting ChatModel<()> for delegated OpenAiModel capability methods. Focused tests pin request model selection, managed usage recovery (including cached-token precedence and absent/zero billing distinctions), readiness behavior, auth handling, and catalog discovery.

Validation

• cargo fmt --manifest-path crates/tinyinference-llm/Cargo.toml -- --check and git diff --check pass.
• Hosted CI passed on final head e5092f5: both stable jobs (format, Clippy, build, all-feature tests, default-feature tests, docs), both Rust 1.88 MSRV jobs, and both supply-chain jobs. The prior PR feat: add TinyHumans managed model adapter #18 ran on the original implementation head and failed stable/MSRV compilation with E0283; this PR carries the compile fix and billing-semantics correction. No green result from that earlier PR is claimed.
• Local package tests passed: 771 unit tests, 2 registry integration tests, 4 tool-validation tests, and 1 doctest. All-feature package check and strict all-target/all-feature Clippy (-D warnings) passed on the same final head. These ran through OpenHuman’s scripts/ci-cancel-aware.sh with CARGO_BUILD_JOBS=1, selecting vendor/tinyagents/vendor/tinyinference/crates/tinyinference-llm/Cargo.toml; full test, check, and Clippy logs are retained in the epic evidence ledger.

History note

PR #18 was marked merged by GitHub after its head commit was accidentally pushed to this fork's epic base ref. No merge button or merge command was used. The epic base now contains that implementation commit; this corrective PR adds the follow-up semantics correction and regression coverage and is the review surface for the final combined head.

Refs #8

Part of tinyhumansai/openhuman#8092