Accept standard reasoning_effort / output_config.effort as effort source

[Blue-B]

What

Read the standard OpenAI reasoning_effort body field (and Anthropic-style output_config.effort) as an effort source, in addition to the existing effort / x-opencode-effort.

Why

OpenAI-compatible clients (e.g. pi, and the OpenAI SDK generally) send the reasoning level as reasoning_effort in the request body. Meridian only read effortHeader (x-opencode-effort) and body.effort, so the effort coming from standard clients was silently dropped and never forwarded to the SDK — every request effectively ran at the model default regardless of the client's chosen level.

Change

src/proxy/server.ts:

const effort = effortHeader
  || body.effort
  || body.reasoning_effort        // standard OpenAI field
  || body.output_config?.effort   // Anthropic-style nesting
  || undefined

Notes

• Purely additive and backward compatible: existing effort / x-opencode-effort keep priority; the new fields are fallbacks.
• body is any (from c.req.json()), so no type changes are required.
• Verified end-to-end: with this change, a client sending reasoning_effort: "xhigh" (or max) reaches the Claude Code SDK --effort flag (confirmed by the SDK validating the value: it must be one of: low, medium, high, xhigh, max).