Workaround for caching in TrackStateProvider

The cache in the TrackStateProvider is not compatible with the new scheduler as it relies on incidents for clearing. In practice we get cache collisions, which interferes with things like calculating IP cuts and hence blocks any meaningful work on selections. See LHCBPS-1835 for more details.

This here is a temporary workaround, and as described in the JIRA task it might leads to other issues for code that keeps pointers to the cached objects (like DecayTreeFitter). For this reason, the clearing of the cache is guarded by a new property UnsafeClearCache, which defaults to false.

The proper fix is to remove the caching and should hopefully come soon.