sql/stats: support per-statement histogram bucket count in CREATE STATISTICS

Currently the histogram bucket count is configurable via the cluster setting `sql.stats.histogram_buckets.count` and the per-table storage parameter `sql_stats_histogram_buckets_count`, but there's no way to override it for a single `CREATE STATISTICS` invocation. This means collecting a one-off high-resolution histogram for diagnostics requires temporarily altering the table's storage parameter and reverting it afterward.

Propose adding an option like `CREATE STATISTICS ... WITH OPTIONS (histogram_buckets = N)` to allow per-statement overrides.

Both `mergeHistograms` (`pkg/sql/stats/merge.go`) and the forecasting system (`pkg/sql/stats/forecast.go`, which operates in quantile-function space) are bucket-count-agnostic, so mixed bucket counts in a column's history are handled gracefully.

Epic: none

Jira issue: CRDB-65478

Start by tracing the CREATE STATISTICS implementation and how it reads sql.stats.histogram_buckets.count and sql_stats_histogram_buckets_count. Review pkg/sql/stats/merge.go and pkg/sql/stats/forecast.go to confirm their bucket-count-agnostic behavior. Done means a single CREATE STATISTICS invocation can override the histogram bucket count without changing the cluster setting or table storage parameter.

Written by the indexing model from the issue text.