Add DeepSeek Pro and MiniMax preview model choices

[thisguymartin]

Closes #5

What changed

Setup previously allowed only one gateway model per provider. Add deepseek-v4-pro and MiniMax-M3.1-Flash-Preview alongside the existing Flash and M3 choices, with independent effort selections and probes. Preserve existing assignments and provider routing.

Update matrix invariants and runner regression tests for both parent routes, including rejection of substituted models. Document MiniMax preview Token Plan access and model-specific thinking behavior.

Verification

• Bun tests, strict typecheck, static invariants, and plugin validation pass.
• The exact candidate is installed in every affected harness.
• The changed behavior passes from each real user surface.
• The installed version, action, and observed result appear below.

195 tests passed. Manifest parsing, claude plugin validate, and git diff --check also passed.

Live evidence:

From this Codex session, the candidate external runner completed four read-only synthetic probes with Claude Code 2.1.283: DeepSeek Flash, DeepSeek V4 Pro, MiniMax M3, and MiniMax M3.1 Flash Preview. All returned the expected marker and reported the requested model with modelVerified: true. See probe evidence for results and limitations. No credentials are included.

The packaged version is 1.4.1, but this candidate has not been installed as a plugin. Installed /setup-pstack behavior in both Claude Code and Codex remains untested. Runner probes do not satisfy that gate or prove hidden reasoning depth.

This PR remains a draft until the exact candidate passes the installed user-surface tests. Do not merge, tag, release, or roll it out.