[octo-patch]

Reason: Correct MiniMax model usage telemetry to apply the supplied per-model rates.

Changes:

• Adds model-specific input, output, and cache-read pricing for MiniMax-M3 and MiniMax-M2.7.
• Records cache-write tokens and charges them using the MiniMax-M2.7 cache-write rate.
• Keeps existing DeepSeek pricing behavior unchanged.

Checks:

• git diff --check
• go test ./...

[octo-patch]

I refreshed the telemetry table to MiniMax-M3.1-Flash and MiniMax-M3 pricing, including the MiniMax-M3 long-context tier, and removed the stale MiniMax-M2.7 cache-write charge. I ran go test ./....