Forward tracking: Try tiled non-exhaustive seeding

Seeding is currently done with two chunks of 32 hits, trying them in a tiled manner of size 16. Let's say we have arrays A and B of size 32:

• A[0:15] x B[0:15]
• A[16:31] x B[0:15]
• A[0:15] x B[16:31]
• A[16:31] x B[16:31]

If they were populated in increasing order (right now it's not the case), then probably A[16:31] x B[0:15] or A[0:15] x B[16:31] would not lead to anything useful.

I suspect we can change the arrays to hold 64 hits instead of 32, and then just process the following:

• A[0:15] x B[0:15]
• A[16:31] x B[16:31]
• A[32:47] x B[32:47]
• A[48:63] x B[48:63]

We would essentially process the same amount of hits as of now (although with less locality), and probably get a higher chance to get the good chi2 candidates.