fix: address review findings (cache path, docs)

- Composite action: cache to runner.temp instead of /usr/local/bin
  (avoids permission issues on runners)
- Document that temperature=0 means server default (omitted from request)
- Note: strconv import already exists (false positive from GPT-5)

llm/client.go

@@ -29,6 +29,8 @@ func NewClient(baseURL, apiKey, model string) *Client {
 }
 
 // WithTemperature sets the temperature for LLM requests.
+// A value of 0 (the zero value) means the field is omitted from the request,
+// causing the server to use its default temperature.
 // If not set (zero value), the server default is used.
 func (c *Client) WithTemperature(t float64) *Client {
 	c.Temperature = t