docs: document function cold start behavior and suggest warm-up patterns

[MajorTal]

Problem

First invocation of an edge function after deploy takes ~2.5s (Lambda cold start). Subsequent invocations are ~1s. For user-facing functions like on-signup, this means a noticeable delay on first login.

Not a bug — Lambda cold starts are expected. But it's not documented, so developers may think something is broken.

Suggested Solution

• Document cold start behavior in llms-cli.txt under the functions section
• Consider a warm-up option (periodic ping) for functions on paid tiers
• Or document the pattern: "For latency-sensitive functions, invoke them once after deploy to warm the container"

Context

Wild Lychee on-signup function: 2.5s on first call, 1s on subsequent calls. User waits 2-3 seconds after first login for their member record to be created.