feat(providers): add MiniMax provider (MiniMax-M2.7, M2.5, 204K context)

[octo-patch]

Summary

This PR adds a providers/minimax package that makes it easy to use MiniMax language models with the Agency library.

MiniMax exposes an OpenAI-compatible REST API (https://api.minimax.io/v1), so the new package is a thin wrapper around the existing providers/openai package.

What changed

File	Description
providers/minimax/provider.go	New provider: New(), model constants, TextToText(), TextToStream(), temperature clamping
providers/minimax/provider_test.go	10 unit tests (temperature clamping, operation creation, constants)
providers/minimax/provider_integration_test.go	1 live integration test (skipped when MINIMAX_API_KEY is unset)
examples/minimax/main.go	Minimal interactive chat loop using MiniMax-M2.7
README.md	Mentions the new provider; marks roadmap item as done

Supported models

Constant	Model name	Context
ModelM27	MiniMax-M2.7	204 K
ModelM27HighSpeed	MiniMax-M2.7-highspeed	204 K
ModelM25	MiniMax-M2.5	204 K
ModelM25HighSpeed	MiniMax-M2.5-highspeed	204 K

Temperature handling

MiniMax requires temperature to be in the range (0, 1]. The provider automatically clamps any nil/zero/negative value to 0.01 and any value above 1 to 1.0, so callers never need to worry about invalid values.

Usage

import (
    "github.com/neurocult/agency/providers/minimax"
)

assistant := minimax.
    New(minimax.Params{Key: os.Getenv("MINIMAX_API_KEY")}).
    TextToText(minimax.TextToTextParams{Model: minimax.ModelM27}).
    SetPrompt("You are a helpful assistant.")

Tests

ok  github.com/neurocult/agency/providers/minimax  (10 unit + 1 integration tests)