Use Cranelift bitcast instead of store & load to bitcast between vectors and non-vectors #4004

Job	Run time
test (ubuntu-latest, x86_64-unknown-linux-gnu)	18m 30s
test_llvm	1m 36s
rustfmt	15s
bench	7m 27s
test (ubuntu-24.04-arm, aarch64-unknown-linux-gnu)	10m 29s
test (macos-latest, aarch64-apple-darwin)	11m 6s
test (ubuntu-latest, s390x-unknown-linux-gnu, gcc-s390x-linux-gnu qemu-user)	5m 27s
test (macos-13, x86_64-apple-darwin)	18m 56s
test (windows-latest, x86_64-pc-windows-gnu)	2m 31s
test (windows-latest, x86_64-pc-windows-msvc)	3m 43s
dist (macos-13, x86_64-apple-darwin)	2m 24s
test (ubuntu-latest, riscv64gc-unknown-linux-gnu, gcc-riscv64-linux-gnu qemu-user)	6m 17s
dist (ubuntu-24.04-arm, aarch64-unknown-linux-gnu)	1m 55s
dist (ubuntu-22.04, x86_64-unknown-linux-gnu)	1m 37s
dist (windows-latest, x86_64-pc-windows-msvc)	2m 15s
dist (macos-latest, aarch64-apple-darwin)	1m 40s
dist (windows-latest, x86_64-pc-windows-gnu)	3m 10s
release	0s
	1h 39m 18s

Provide feedback