llama-swap only waited a maximum of 5 seconds for an upstream
HTTP server to be available. If it took longer than that it will error
out the request. Now it will wait up to the configured healthCheckTimeout
or the upstream process unexpectedly exits.
Replace previously hardcoded value for `/health` to check when the
server became ready to serve traffic. With this the server can support
any server that provides an an OpenAI compatible inference endpoint.
