rgw: add accurate Retry-After header to rate-limit responses

rgw: add accurate Retry-After header to rate-limit responses

When RGW rejects an S3 request with -ERR_RATE_LIMITED, return an HTTP
Retry-After header that tells the client how many seconds to wait before
retrying. The value is computed from the current token-bucket deficit and
replenishment rate, so a polite client that honors the header will retry
right around when its tokens recover.

The delay is threaded out of the rate limiter through req_state and onto
the response in abort_early(). The rate-limit check now returns int64_t
seconds (0 = allowed) instead of bool, with a 1-second floor on the
rejected path.

The seconds-to-recover calculation is factored into a small static helper
RateLimiterEntry::compute_delay(limit, needed, interval) so it can be
unit-tested directly. Each dimension (ops, bytes) calls it independently
and the caller takes the max, leaving room for additional limit
dimensions in the future.

The user-facing documentation is updated to describe the new header and
to suggest proxy-layer remapping for operators who need 429 instead of
the existing 503.

Tracker: https://tracker.ceph.com/issues/64083
Tracker: https://tracker.ceph.com/issues/67655
Signed-off-by: Lumir Sliva <61183145+lumir-sliva@users.noreply.github.com>