Remove canonicalize_identity from A2B and B2A parsing

This should be a post-parse optimization, not a part of parsing. Having
this be part of parsing is making testing our Rust-based parser more
complicated.

Also, while I'm complaining about this, if we're going to do this sort
of thing, we should be doing the full parameter fit (cause we often
encounter non-linear parametric curves that have been stuffed into 1D
LUTs in the r/g/bTRC tags).

Update test profile expectations to match.

Bug: 452666425
TAG=agy
CONV=c6043f3d-bcc5-4d8b-b1d3-253bc2e6c1c2
Change-Id: I7de3d80355e17d1d6b1c3357d484a3c7e3caec22
Reviewed-on: https://skia-review.googlesource.com/c/skcms/+/1344916
Reviewed-by: Florin Malita <fmalita@google.com>
Commit-Queue: Christopher Cameron <ccameron@google.com>