Write surface bounds once more.

In ATLASRECTS-6858, we realised that we had been writing out surfaces and silently dropping their bounds. For some, e.g. CylinderSurface, this makes them invalid and unusable when they are read back in.

I experimented with several possible fixes, and after discussion in RIG, decided that we simply had to revisit our 10 year old decision to ignore bounds.

In this MR I attempt to restore the BoundSurfaceCnv, create a new BoundSurface_p2 (because it now inherits from Surface_p2 and increment the TrackCollection top level object to hold the new collections.

This seems to have a pretty negligible increase in the ESD size, from the 25 event test.

Taking an unslimmed InDet TrackCollection, the size changes as follows:

   26410.442 kb     5538.593 kb      221.544 kb        0.000       25  (B) Trk::TrackCollection_tlp7_CombinedInDetTracks

from the original:

   26283.019 kb     5397.960 kb      215.918 kb        0.000       25  (B) Trk::TrackCollection_tlp6_CombinedInDetTracks

The total 25 event ESD changes from 26Mb to 27Mb.

With thanks to @mnowak for all his help.

Ping also @goetz and @mhodgkin