feat(prime): ACP model and reasoning effort pickers (#2455) assessment

[rynfar]

Background & Upstream Change

Upstream Prime Agent PR #2455 added support for model and reasoning effort selection over the Agent Client Protocol (ACP) surface:

• Allows ACP clients to query available models and their supported reasoning effort levels (e.g. low, medium, high, max, or off).
• Allows clients to pass selected model and effort parameters in ACP session initialization and prompt requests.

Analysis for Pylon

1. Pylon ACP Architecture

• Pylon already provides an ACP adapter (apps/server/src/acp) that bridges ACP clients to Pylon's backend provider instances.
• Pylon's provider abstractions already understand model configurations and effort/thinking parameters across supported providers (Anthropic, OpenAI, Prime, Gemini).

2. Assessment of Need

• Priority: Low.
• Upstream #2455 is primarily useful when Prime Agent acts as an ACP server directly serving external ACP clients (such as Zed or Neovim plugins).
• Pylon's internal orchestration communicates with Prime Agent primarily via the private daemon socket protocol (prime-agent.daemon), where session configuration and model selection already pass through the daemon adapter.
• For users running Pylon as an ACP server to external clients, exposing reasoning effort selection in Pylon's ACP protocol handlers could improve parity, but is not blocking core developer workflows today.

Recommendation

• Track as a low-priority enhancement.
• If upstream's daemon bridge exposes effort picker metadata in the model catalog, verify that Pylon's catalog ingestion parses the effort options without schema errors.