feat: make _canonical_matrix faster #16604

Job	Run time
doctest (1.10, ubuntu-latest)	25m 57s
doctest (1.11-nightly, ubuntu-latest)	25m 41s
test (1.6, short, ubuntu-latest)	1h 23m 58s
doctest (nightly, ubuntu-latest)	28m 28s
test (1.6, long, ubuntu-latest)	1h 1m 26s
doctest (1.10, ubuntu-latest, depwarn=error)	27m 4s
test (1.10, short, ubuntu-latest)	1h 10m 23s
doctest (1.10, macOS, RPTU)	12m 56s
test (1.10, long, ubuntu-latest)	45m 41s
test (1.11-nightly, short, ubuntu-latest)	1h 22m 18s
test (1.11-nightly, long, ubuntu-latest)	54m 24s
test (nightly, short, ubuntu-latest)	1h 20m 10s
test (nightly, long, ubuntu-latest)	52m 1s
test (1.10, short, ubuntu-latest, depwarn=error)	1h 10m 42s
test (1.10, long, ubuntu-latest, depwarn=error)	45m 28s
test (1.10, short, macOS, RPTU)	13m 53s
test (1.10, long, macOS, RPTU)	14m 48s
test (1.10, book, ubuntu-latest)	52m 23s
	14h 7m 41s

Provide feedback