produce message supports rate limit in fail-fast

**Is your feature request related to a problem? Please describe.**

now if one topic was rate limited, we will use disableCnxAutoRead, which will not read data from buffer.
one topic sends a message in t1, we disabled then, and in t2, we start to read, and processed (maybe t2-t1 > the timeout of users)

but when we processed in t3, when we send result to producer, it is timeout.

**Describe the solution you'd like**

we can choose the rate limit stratery, failfast or disableautoread

**Describe alternatives you've considered**

implement the fail fast mode, when rate limited, we will send servererror to producer

**Additional context**
Add any other context or screenshots about the feature request here.

when we use disableCnxAutoRead , sometimes ,if producers close the connection by close producer, but we cannot read the request, they we cannot close the connection, there will be many close-wait in broker

The issue names no files, tests, or entry points. Start by tracing broker rate-limit handling, disableCnxAutoRead, and the producer response path; determine where a fail-fast strategy can be selected and how a rate-limited request should return a server error. Done means the selected strategy avoids delayed producer timeouts and connections can close cleanly.

Written by the indexing model from the issue text.