Bug fix visual sort voting logic

[rbezentrigo]

Fixes in Visual Voting Logic

This PR addresses two critical issues in the visual association logic that caused suboptimal assignment of observations to tracks:

1. Fallback Matching for Non-Best Matches

Problem:

In the previous implementation of BestFitVoting, if multiple observations preferred the same track, only the highest-scoring one would be matched. All others were immediately assigned to themselves (fallback), even if they had other suitable — but slightly less optimal — track options.

Fix:

Introduced BestFitVotingWithFallback, which iterates through each observation’s ranked candidates and assigns it to the best available track. Only if none are available, it falls back to self-assignment.

2. Self-Assigned Observations Blocked Positional Voting

Problem:

Observations that failed to match any track via feature distance were assigned to themselves (to fulfill API expectations). These were mistakenly treated as “matched” and were excluded from the positional (geometry-based) assignment phase.

Fix:

Self-assigned observations are now filtered out from feature_winners, allowing them to continue into the positional voting stage. This enables geometry-based matching as a true fallback path.

Combined Result:

• Observations now have a fairer chance of being matched to a valid track, even if their top match is already taken.
• Observations without any strong feature match are no longer prematurely blocked from positional matching.

[bwsw]

@rbezentrigo thank you for your PR, I will review it soon.