[Enhancement] Expose the actual tablet version-limit counter (_approximate_num_rowsets) as a Prometheus metric

### Search before asking

- [x] I had searched in the [issues](https://github.com/apache/doris/issues?q=is%3Aissue) and found no similar issues.

### Description

Writes to a tablet are rejected once its version count passes `max_tablet_version_num`, and it fails with `TOO_MANY_VERSION` (E235). The check reads a counter called [_approximate_num_rowsets](https://github.com/apache/doris/blob/master/be/src/cloud/cloud_rowset_builder.cpp#L204).
The only related metric, `tablet_version_num_distribution`, reads a different counter that updates less frequently and skips tablets that are currently being written to. This hides potential issues until there is a failure.
There's no Prometheus metric for `_approximate_num_rowsets`. The only way to read it is via admin HTTP GET to /api/hotspot/tablet?metrics=num_rowsets.

### Solution

1. Add a metric fed from [_approximate_num_rowsets](https://github.com/apache/doris/blob/master/be/src/cloud/cloud_rowset_builder.cpp#L204) using the same pattern as the existing histogram in doris_metrics.{h,cpp}.
2. Report it for all tablets, including ones under active load

### Are you willing to submit PR?

- [x] Yes I am willing to submit a PR!

### Code of Conduct

- [x] I agree to follow this project's [Code of Conduct](https://www.apache.org/foundation/policies/conduct)

Start with the _approximate_num_rowsets use in be/src/cloud/cloud_rowset_builder.cpp, then read the existing histogram pattern in doris_metrics.h and doris_metrics.cpp. Trace how tablet metrics are reported and confirm the new Prometheus metric includes tablets under active load and reflects the actual counter rather than the existing distribution.

Written by the indexing model from the issue text.