Fixup ford radar

[sshane]

• MRR_Header_InformationDetections->CAN_NUMBER_OF_DET exactly matches our parsed points using CAN_DET_VALID_LEVEL_xx when assuming the scan index signal is a sync signal
• Each track message's CAN_SCAN_INDEX_2LSB_xx signal matches the header's when triggered on the header message
• Every cycle we now receive exactly 65 track updates when triggered on the header message

[sshane]

CAN_NUMBER_OF_DET was always lower by 3-6 times from our number of points, now they match exactly!

It was originally thought that CAN_SCAN_INDEX_2LSB_xx was a sort of multiplexing signal, but it makes more sense that along with the header message it's used to ensure that the latest updates are kept up to sync.

Will have to look into the flickering that was noticed in the original PR (commaai/openpilot#24296) though, perhaps some more robust track switching logic is needed, or maybe it's not a problem since Harald added it after the original PR.

[sshane]

All track message's LSB scan index signals always match the header counter's 2 LSB when waiting for it now, before it would usually have a few messages that mismatched every cycle (so we should have waited a bit longer for the header message to have the most up to date detection/track messages)

[sshane]

I colored scan indexes of 0 and 2 as green, and 1 and 3 as purple. They're grouped into close range tracks and far away/less accurate tracks.

CAN_SCAN_INDEX and CAN_LOOK_INDEX always match (16 bit numbers). And CAN_LOOK_ID always follows this mapping to the 2 LSB of the indexes: ID 0 = INDEX 1, ID 1 = INDEX 3, ID 2 = INDEX 0, and ID 3 = INDEX 2. This means we can use any one of these signals to group the points, and the original logic was correct.

ford-radar-sorted-by-scan-index.mp4

This is very clearly seen here:

View

[sshane]

I then filtered for only CAN_DET_SUPER_RES_TARGET_xx=1 tracks. Fewer points, and they don't capture obvious objects like the lead.

ford-radar-only-super-res.mp4

[sshane]

I also just realized the groups are explainable with these two signals:

I had some false tracks on the road with the long-range detection mode, probably can filter them out with this:

Here is a line at ~45m:

ford-radar-40m.mp4

And filtered out long range points < 35m:

[sshane]

mostly in