feat: support extra field in chatcompletion tool_calls for gemini openai compat

[bkvarda]

Summary

Various model providers have started including additional fields in in tool_calls of openai-compatible(ish) chatcompletion responses, for example gemini's thought_signature and databricks' thoughtSignature. This change adds an extra Map field in DeltaToolCallFunction to support arbitrary additional fields. Then added inclusion of these additional fields in databricks format_messages which fixes 400: content block is missing a thought_signature errors when trying to use tools w/ gemini-3 in databricks.

Type of Change

• Feature
• Bug fix
• Refactor / Code quality
• Performance improvement
• Documentation
• Tests
• Security fix
• Build / Release
• Other (specify below)

AI Assistance

• This PR was created or reviewed with AI assistance

Testing

Added unit tests and tested manually with databricks-gemini-3-pro and databricks-gemini-3-flash

Screenshots/Demos (for UX changes)

[Copilot]

Pull request overview

This PR adds support for arbitrary extra fields in OpenAI-compatible chat completion tool calls to handle provider-specific metadata like Gemini's thought_signature and Databricks' thoughtSignature. This fixes 400 errors when using tools with gemini-3 models through Databricks.

• Adds an extra field with #[serde(flatten)] to DeltaToolCall to capture non-standard fields
• Updates message formatting to include metadata fields in outgoing tool call requests
• Extracts and preserves extra fields from responses in both streaming and non-streaming paths

Reviewed changes

Copilot reviewed 3 out of 3 changed files in this pull request and generated no comments.

File	Description
crates/goose/src/providers/openai.rs	Reorganizes imports (moved to top of file)
crates/goose/src/providers/formats/openai.rs	Adds extra field to DeltaToolCall, updates format_messages to include metadata in tool calls, extracts extra fields in response_to_message and streaming handler, adds comprehensive tests
crates/goose/src/providers/formats/databricks.rs	Updates format_messages to include metadata in tool calls (mirrors OpenAI changes), adds tests for metadata handling

[Copilot]

Pull request overview

Copilot reviewed 3 out of 3 changed files in this pull request and generated no new comments.

[DOsinga]

we just merged: #6035 which does something similar - can you check whether that covers your intentions? it feels like it is doing both more (handling the thought signatures) and also less (doesn't store all meta information)

[bkvarda]

@DOsinga I tried with #6035 but it didn't seem to resolve scenarios in which the goose provider is parsing responses as OpenAI-compatible responses using stream_openai_compat which call the oai response_to_streaming_message (like the databricks provider does, but a number of others look like they do such as openrouter and xai).

brandon.kvarda@HQ2MPQV7RY goose % git checkout fc8a4b4adfbd771c09d867c0a5c04d643f83905a
HEAD is now at fc8a4b4adf fix(google): Fix 400 Bad Request error with Gemini 3 thought signatures (#6035)
brandon.kvarda@HQ2MPQV7RY goose % cargo build
    Finished `dev` profile [unoptimized + debuginfo] target(s) in 1.08s
brandon.kvarda@HQ2MPQV7RY goose % GOOSE_MODEL=databricks-gemini-3-pro DATABRICKS_HOST=https://e2-demo-field-eng.cloud.databricks.com ./target/debug/goose
starting session | provider: databricks model: databricks-gemini-3-pro
    session id: 20251222_14
    working directory: /Users/brandon.kvarda/othercode/goose

goose is running! Enter your instructions, or try asking what goose can do.

Context: ○○○○○○○○○○ 0% (0/128000 tokens)
( O)> read a file in this dir

─── read_module | code_execution ──────────────────────────
module_path: developer

Ran into this error: Request failed: Bad request (400): {
  "error": {
    "code": 400,
    "message": "Unable to submit request because function call `default_api:code_execution__read_module` in the 2. content block is missing a `thought_signature`. Learn more: https://docs.cloud.google.com/vertex-ai/generative-ai/docs/thought-signatures",
    "status": "INVALID_ARGUMENT"
  }
}
.

Please retry if you think this is a transient or recoverable error.

⏱️  Elapsed time: 5.31s

[spazbg]

I can validate that in the latest release 1.18.0 this issue is still present, so please share a fix ASAP. We want to use this dirt cheap and fast Gemini 3 flash model :)

[bkvarda]

@spazbg which provider are you using? if not databricks any chance you could build goose cli from this and validate if it works for your provider as well? EDIT: Also you might have to do something like this in the specific provider also.