Fix SaveCurrentTrackTask GPX write and error reporting

[dmvkosm]

Summary

• Write each GpxFile from collectRecordedData() instead of always writing the task constructor instance
• Propagate write failures to SaveGpxListener.onSaveGpxFinished(Exception)
• Use mkdirs() for the target directory

Related issues

#25076 (silent save failure when finishing track recording)

Test plan

• Finish recording a track with storage available — file appears
• Simulate write failure (read-only dir) — listener receives error, path not cleared

[Chumva]

This fixes the core bug of writing the wrong GpxFile, but it breaks existing callback users. Several callers use the original gpxFile.getPath() after onSaveGpxFinished() to share/export the saved current track. With this change, only the collected per-file gpxFile gets its path set by writeGpxFile(), while the constructor gpx can remain empty, so callers may end up with new File("").

The task should return/set the actual saved output file path for the callback flow, or the callers need to be updated to use the saved file(s). Also if (error == null && gpx.getPath().isEmpty()) gpx.setPath("") is a no-op now

[dmvkosm]

Updated: SaveResult now sets gpx.setPath(savedPath) before onSaveGpxFinished(), then clears path only for cache-dir saves after the callback (share/export flow).

[Chumva]

The error propagation is a good fix, but this version introduces a couple of regressions.

collectRecordedData() can return multiple date-split GPX files, while the task stores only the last savedPath. Export/share callers read only gpx.getPath(), so multi-day recordings may export only the last saved part.

Also, when the input gpx already has a path, this can replace it with tracksDir/<date>.gpx, changing the caller’s original path unexpectedly.

I think this task should either write the passed GpxFile once for export/share, or delegate full recorded-track persistence to saveDataToGpx() instead of partially duplicating it.