[octo-patch]

Reason: Make MiniMax M3.1 Flash the ordinary default while preserving explicit legacy and custom model requests.

Summary

• Make MiniMax-M3.1-Flash the MiniMax provider default and first supported model choice.
• Keep explicit MiniMax-M3, MiniMax-M2.7, saved model IDs, and custom IDs available without reset.
• Apply Flash's one-million-token context metadata, adaptive thinking, default max effort, and outgoing reasoning_effort serialization.
• Keep existing M3 attachment routing and non-MiniMax behavior unchanged.

Validation

• git diff --check
• go test ./internal/defaults ./internal/llm/minimax (cached, offline)
• Mocked HTTP boundary test added for Flash adaptive thinking, effort, model, and omission of reasoning_split.
• Full go test ./...: local dependency cache is incomplete and network access timed out while fetching pinned modules; remote CI is running.
• Manual CLI/TUI check: unavailable in the credential-free worker environment.

User-visible impact

MiniMax starts on M3.1 Flash. Explicit saved or requested M3, M2.7, legacy, and custom model IDs continue through the existing provider and attachment/history paths.

Breaking changes

None. The default MiniMax model changes for new/default selections; explicit model selections are preserved.

Notes

The repository's upstream default branch now carries a maintenance notice. This PR remains open for maintainer review. No paid inference or provider credentials were used; the provider test uses a local mocked HTTP server.

[shayne-snap]

Hi @octo-patch. Thanks for the PR. Unfortunately, this repository is no longer maintained.

I'm gonna close this PR.