Cleanup state location and track type usage in track v3 and converters

Unifies state enum definitions and removes type conversion gymnastics now unnecessary. There is one notable change in behaviour of FittedForward vs. Long tracks. The former were silently converted to the latter by the track_v3 to track_v1 converter. This only affects baseline code tests because the FittedForward type stems from HLT1 on CPUs.

I believe it is much less error-prone to keep them separated. An alternative would be to get rid of this type completely and simply treat it as standard Long track (as was done before).

Based on !4370 (closed) (commits contained here)

Goes with Rec!3677 (merged) and Alignment!432 (merged)

Resolves #335 (closed)