feat: add The Grid provider

[BarneyChambers]

Summary

Adds The Grid as an OpenAI-compatible provider with 9 models: internal/providers/configs/thegrid.json, its registration in internal/providers/providers.go, and the InferenceProvider constant in pkg/catwalk/provider.go — the same three-file shape as #373.

The Grid is an OpenAI-compatible inference platform (docs, API base https://api.thegrid.ai/v1). These records are generated from the same source of truth that drives our live /v1/models endpoint, so the values in this PR match what the API enforces.

How these values are derived

Grid models differ from single-model providers in ways that affect the values here:

• Each model id routes across a pool of underlying models rather than naming one. Published limits are the guaranteed floor across that pool — the minimum every pooled model accepts — so requests within the published limits never fail on limit grounds.
• The max input limit is a real, independently enforced cap. It is not derivable as context − max output.
• Pricing is uniform across token types: input, output, and cached tokens bill at the same rate, so cost_per_1m_in, cost_per_1m_out, and both cached figures are equal by design rather than by omission. The published figure is the trailing 7-day volume-weighted average of the live spot price; we refresh it whenever we update these records.
• type is openai-compat and api_key is $THEGRID_API_KEY, matching how the other third-party OpenAI-compatible providers in configs/ declare themselves.
• supports_attachments is true only on the *-max tier, which is what architecture.input_modalities reports in our catalog; the prime and standard tiers are text-only.

Models

model	context	max input	max output	USD / 1M tokens (in = out)
agent-max	1,000,000	922,000	128,000	1.5885
agent-prime	196,608	120,000	30,000	0.1422
agent-standard	128,000	120,000	16,000	0.02
code-max	1,000,000	922,000	128,000	1.7553
code-prime	196,608	120,000	30,000	0.1493
code-standard	128,000	120,000	16,000	0.0288
text-max	1,000,000	922,000	128,000	1.68
text-prime	196,608	120,000	30,000	0.1017
text-standard	128,000	120,000	16,000	0.03

default_large_model_id is agent-max and default_small_model_id is text-standard.

Verification

• go build ./..., go vet ./..., and go test ./... all pass on this branch (Go 1.26), and gofmt -l ./internal ./pkg is clean.
• TestValidDefaultModels/The_Grid passes both default model ids resolve within the models array.
• Confirmed the embedded config decodes into catwalk.Provider with every field populated (no JSON key silently missing a struct tag): type=openai-compat, endpoint=https://api.thegrid.ai/v1, key=$THEGRID_API_KEY, 9 models with limits, prices, and capability flags all set.
• Live values are checkable without auth: curl -sS https://api.thegrid.ai/v1/models.

These entries are maintained by an automated generator with daily reconciliation against the published registry, so drift is caught and corrected promptly. Happy to add a cmd/thegrid generator and wire it into the auto-update workflow (the #463 shape) if you'd prefer these refreshed automatically rather than by PR.

[BarneyChambers]

Please let me know if you have any style-guides for submitting pull requests and I would be happy to amend my request to align with that. Thanks for reading!