Physical record-count refresh

For ordinary DBFs, a physical count refresh derives the row count from the file body length, matching Harbour DBFCDX’s file-size rule. A stale high header no longer resurrects rows after a peer shrinks or empties the body; a stale low header does not hide complete rows already on disk. Optional EOF bytes are excluded by integer division for ordinary record lengths.

Full-table encryption retains fixed-length records. Partial encryption has a trailing bitmap and keeps its existing header-count path instead. This patch does not change either encryption format.

The driver refresh result now reaches local AdsGetRecordCount and the server’s GetRecordCount reply. A failed read/size query or truncated DBF returns an error, not a successful stale count. Local caller output is not assigned on refresh failure. NTX, ADT and cache wrappers forward the result. Existing navigation refresh calls are not broadened into a new error policy.

No remote count-cache policy, OPENADS_FRESH_COUNTS switch, ordered-position optimization, protocol change or Harbour RDD-core edit is included. A client cache hit still follows the existing policy and may miss a peer append; this patch corrects the value and error result when a physical refresh occurs.

This work is carried in Pritpal Bedi’s mtfix41 fork release. The upstream regressions cover stale headers, peer append/shrink, empty/EOF-only bodies, truncation errors on local and direct-wire paths, and encrypted counts.