Run with PSD and segments

This should fix our PSD and segment issues.

There's a couple of lingering issues before it should be merged:

• It's not outputting coincs (possibly related to PSD?)
• The corresponding GstLAL commit should be cleaned up and merged first: spiir-group/gstlal!2 (merged)
• I should go through an enumerate all the changes in this MR.

There's a good few changes. The main things are:

• Fixing a typo in multiratespiir.c
  • Subbuffers are now created with offset = offset * bps, which Apparently is necessary. I haven't tried to understand it yet, but it's something I'd undone because it looked wrong (I think) and then we got all sorts of invalid memory issues & infinite SNRs
• Postcoh now allocates & copies memory from the adapter without relying on a modified core gstreamer function.
• I've tried to simplify any type changes which I made before (and there were some clear mistakes). So types should only have changed from 32 bit to 64 bit, or signed to unsigned, as absolutely necessary.
• I've similarly tried to simplify memory mapping & unmapping, so I'm only doing that as necessary (And I think there was a fix in filesink, where previously I was unmapping memory too early. Didn't get any clear test issues on it though.)
• There's a few extra sanity checks/debug steps which should be removed before any serious testing, but are harmless for this MR
• We changed a bunch of async cuda functions to sync just to be safe. In general we should use async if it's safe to do so, but for debug it's a nice sanity check to use sync.

Tests

I've run 8000 second injection tests here: /fred/oz996/tdavies/spiir_project/sources/testing/injection_tests/local/run1
Note there's no coincs, turns out this is normal for gstreamer_python_upgrade, fars are all 0...

The zerolags otherwise look good to me (minor differences across the board, but only minor).

There's ongoing issues, but more test discussion in spiir-group/gstlal!2 (merged)