refactor(core)!: one tool-call and one tool-result type per protocol layer

[cunninghamcard-bit]

What

Unify tool calls and provider-executed tool results in aimux-core so providers, recording, replay and bindings use shared payload structs instead of repeating field lists. This branch is based on origin/rfc-0036/integration and precedes output-content-types in the public-surface alignment stack. It separates raw provider argument text from parsed call-layer input, makes provider tool choice optional, and removes the self-generated contract data rather than treating Rust's former serialization as the upstream contract.

Before and after

Type or behavior	Before	After and upstream reference
Provider tool calls	Generation and stream variants repeated fields; stream input mixed raw text and parsed JSON.	GenerateContent::ToolCall(RawToolCall) and StreamPart::ToolCall(RawToolCall) carry input: String, corresponding to LanguageModelV4ToolCall.
Call-layer tool calls	Separate raw-call definition and conversion heuristics.	ToolCall carries parsed Value, or the original argument string when parsing fails, with invalid and error; TypedToolCall is the upstream reference. TextStreamPart is currently an alias for StreamPart<ToolCall>.
Provider-executed results	An unused standalone result struct and repeated enum payloads.	GenerateContent::ToolResult and StreamPart::ToolResult share ToolResult, including required tool_name and optional dynamic and provider_metadata, corresponding to LanguageModelV4ToolResult.
Tool choice	CallOptions always supplied a choice.	CallOptions.tool_choice is optional; direct provider calls omit an unset choice. generate_text and stream_text supply auto when a nonempty tool list is supplied without a choice, following ai/src/prompt/prepare-tool-choice.ts.
Tool examples and arguments	Examples were arbitrary JSON values; provider arguments could be nonobjects.	FunctionTool.input_examples contains { input: object }; ProviderTool.args is a JSON object, following LanguageModelV4FunctionTool and LanguageModelV4ProviderTool.
Thought signatures	Tool calls and prompt parts exposed a separate signature field.	Signatures travel through the vendor namespace in provider metadata on output and provider options on input, where the upstream Google and Vertex modules read and write them.
Legacy argument decoding	Generation accepted a nonstring input and serialized it back to text; stream conversion guessed its representation.	The compatibility decoder and raw_tool_input heuristic are deleted. Provider input must be raw text, as in LanguageModelV4ToolCall.

JSONSchema7, dotted provider-tool IDs and NonNullable<JSONValue> constrain TypeScript types without upstream runtime enforcement at these type declarations. Rust keeps Value for schemas and results and String for IDs; this branch adds no corresponding runtime checks. This does not imply that vendor option parsers accept arbitrary values.

Tests

Delete contract-tests/, its self-generated JSON and runner, together with the 16 tests reading that data. These files recorded Rust's own former output rather than upstream payloads. CI, the local CI script and contributor and binding documentation stop invoking or recommending those readers; the generator drift checks remain.

Delete hand-written tests that pinned removed shapes and the invented-data tests in openai_output.rs and moa.rs. Restore the tool-choice cases ported from the upstream OpenAI, Groq, Mistral and Amazon Bedrock suites, adapting them to optional provider choice while preserving their upstream request expectations. Other retained tests follow the shared payload structs. Cassette replays are kept; aimux-providers/tests/cassettes and fixtures/ have no changes in this branch.

The diff also deletes older provider test suites, including suites with upstream-port annotations. The restored tool-choice cases do not establish complete upstream suite coverage; this body makes no claim that every deleted provider case was hand-written or restored.

Not in this pull request

• Provider ToolChoice serialization still uses bare strings for auto, none and required, while LanguageModelV4ToolChoice uses tagged objects. This refactor changes optionality, not that existing wire representation.
• Serde naming remains pending the maintainer's decision: snake_case fields, PascalCase external tags, Option fields serialized as null where omission is not configured, and generated Node types with required-nullable fields. Those binding contract choices are not resolved here.
• The call-layer ModelMessage tool-result shape remains pending the maintainer's decision. TextStreamPart also still carries the provider result rather than the upstream call-layer input/output result and lacks a separate tool-error event for invalid non-provider-executed calls; call-layer result and error separation is outside this payload refactor.
• The public TextStreamPart alias is replaced by a separate enum on provider-result-types, where response metadata is absorbed into the finish step. That later change is not part of this branch.
• Function-tool provider options still use a broad JSON namespace map here; the shared namespace type is introduced on provider-options-types, which owns option and metadata alignment.
• A repaired call that fails validation is still wrapped as a repair error, and invalid-call fallback still uses plain JSON parsing rather than the upstream secure parser. These existing parser differences are not fixed by consolidating the payload types.
• Provider prompt unions are handled on provider-prompt-types; the remaining output content unions are handled on output-content-types, to keep this change focused on tool payloads.

Not verified

Go, Java, Kotlin, Swift and Flutter type and decoder changes were not compiled locally. Node npm test and Python pytest were not run. CI YAML was only syntax-checked; CI execution is not established. This documentation rewrite did not run compilation or tests, and retained upstream test coverage has not been exhaustively mapped.

Verification

Gate on the final commit 8a5acf97: cargo fmt --check, cargo clippy --workspace --all-targets -D warnings, cargo doc, tests of core, provider-utils, providers, ffi, web, replay and cli: 3,321 passed, 0 failed; Node and Python bindings cargo check; provider boundary script; gen_ts_types.py --check, gen_providers_doc.py --check. Cassettes and fixtures/aisdk unchanged relative to the base. 136 files, +1107 / −25133.

🤖 Generated with Claude Code

https://claude.ai/code/session_01WGwanvo9WLU9HWD7sRR8TS