*: forwarding HTTP requests can fail with "net/http: cannot rewind body after connection loss"

## Bug Report

### What did you do?

With audit or rate-limit middleware enabled, send HTTP requests to a PD
follower so that the follower forwards them to the PD leader via
`customReverseProxies` (e.g. `GET /pd/api/v1/cluster`).

When the forwarding `http.Client.Do` needs to retry on a new connection —
typically due to a stale keep-alive connection or HTTP/2 GOAWAY — the retry
fails.

### What did you expect to see?

The forwarded request is transparently retried on a new connection and
succeeds.

### What did you see instead?

The forward fails with:

```
[ERROR] [apiutil.go:465] ["request failed"] [error="[PD:http:ErrSendRequest]send HTTP request failed: Get \"https://.../pd/api/v1/cluster\": net/http: cannot rewind body after connection loss"]
```

Root cause: when audit/rate-limit middleware is enabled,
`requestutil.getBodyParam` replaces `r.Body` with
`io.NopCloser(bytes.NewBuffer(buf))` without setting `r.GetBody`. This turns
what was originally `http.NoBody` into a non-nil body reader with no rewind
function. On the forwarding path, `net/http` Transport probes the body
(`didRead=true`), and on connection loss cannot rewind because `GetBody` is
nil, so the retry path errors out with the message above.

### What version of PD are you using (`pd-server -V`)?

master (reproduces on current master and on the release branches that carry
the same audit middleware code path).

Start by reading requestutil.getBodyParam and the forwarding path around apiutil.go:465, focusing on how audit or rate-limit middleware changes r.Body and how customReverseProxies invokes http.Client.Do. Reproduce with GET /pd/api/v1/cluster through a PD follower and a stale connection or HTTP/2 GOAWAY. Done means the forwarded request can retry on a new connection without the rewind error.

Written by the indexing model from the issue text.