feat: add LLM Gateway kimi-k2.7-code, nemotron-3-ultra-550b, grok-build-0-1

[smakosh]

Adds the newest text models from the LLM Gateway catalog that aren't yet in the repo, using the base_model structure to inherit from the canonical model registry with gateway-specific cost overrides.

Models added (under providers/llmgateway/models/)

Model	base_model	Released	Cost (in/out, $/1M)
kimi-k2.7-code	moonshotai/kimi-k2.7-code	2026-06-12	0.95 / 4 (cache_read 0.19, [interleaved])
nemotron-3-ultra-550b	nvidia/nemotron-3-ultra-550b-a55b	2026-06-04	0.5 / 2.5 (cache_read 0.15)
grok-build-0-1	xai/grok-build-0.1	2026-04-16	1 / 2 (cache_read 0.2, +200k context tier 2/4/0.4)

Notes

• Cost values reflect LLM Gateway's pricing; all other metadata (limits, modalities, family, reasoning) is inherited from the existing registry base specs.
• Image / video / TTS / embedding models from the gateway are intentionally excluded — this PR is text models only.
• bun validate passes; all three resolve correctly.

🤖 Generated with Claude Code

[rekram1-node]

plz add reasoning_options

[smakosh]

Done — added reasoning_options to all three:

• kimi-k2.7-code and grok-build-0-1: [{ type = "effort", values = ["low", "medium", "high"] }] (matching the other kimi/grok gateway models)
• nemotron-3-ultra-550b: [{ type = "toggle" }] per its nvidia source definition

bun validate passes.