Perplexity Agent catalog is out of sync with GET /v1/models

Problem

The perplexity-agent catalog in api.json has 22 entries, while Perplexity's authenticated GET /v1/models returned 45 Agent API model IDs on 2026-08-27. Only 15 IDs overlapped exactly.

Perplexity documents this endpoint as the source for dynamic model discovery:
https://docs.perplexity.ai/api-reference/models-get

IDs returned by Perplexity but missing from models.dev

anthropic/claude-fable-5
anthropic/claude-opus-4-8
anthropic/claude-opus-5
anthropic/claude-sonnet-5
google/gemini-3.1-flash-lite
google/gemini-3.5-flash
google/gemini-3.5-flash-lite
google/gemini-3.6-flash
google/gemini-3.7-flash
openai/gpt-5
openai/gpt-5-nano
openai/gpt-5.3-codex
openai/gpt-5.4-mini
openai/gpt-5.4-nano
openai/gpt-5.6-luna
openai/gpt-5.6-sol
openai/gpt-5.6-terra
perplexity/deepseek-v4-flash-0731
perplexity/deepseek-v4-pro-0813
perplexity/glm-5.2
perplexity/glm-5.3
perplexity/kimi-k2.7-code
perplexity/kimi-k3
perplexity/nemotron-3-ultra-550b-a55b
perplexity/nemotron-3.5-lightning-30b-a3b
xai/grok-4.20-multi-agent
xai/grok-4.20-non-reasoning
xai/grok-4.20-reasoning
xai/grok-4.3
xai/grok-4.5

IDs in models.dev but not returned by the endpoint

deepseek/deepseek-v4-flash-0731
google/gemini-2.5-flash
google/gemini-2.5-pro
moonshot-ai/kimi-k2.7-code
moonshot-ai/kimi-k3
nvidia/nemotron-3-super-120b-a12b
xai/grok-4-1-fast-non-reasoning

Several missing/stale pairs appear to be namespace changes, for example moonshot-ai/kimi-k3 versus the currently callable perplexity/kimi-k3. Exact IDs matter because the models.dev README says consumers should use the Model ID for lookup, and Perplexity accepts the IDs returned by this endpoint.

The separate perplexity catalog also omits the two embedding models, but that appears to be part of the schema limitation tracked in #572.

Start by inspecting the perplexity-agent catalog in api.json and compare its model IDs with the authenticated GET /v1/models response documented by Perplexity. Update the catalog so the callable Agent API IDs and namespaces match the endpoint, while leaving the separately noted embedding-model limitation unchanged; done means the catalog reflects the verified endpoint data exactly.