Use 64-bit implementation for Fp32

[armfazh]

Using 64-bit implementation for Fp32, previously, it was using the 128-bit implementation.

Another option is to use directly a dedicated 32-bit implementation, but I will propose that in PR #1098 .

$ cargo bench --bench speed_tests -- poly_mul

poly_mul	n	Time	Change
fft	1	210.72 ns	-19.316%
fft	30	5.1285 µs	-39.153%
fft	60	11.299 µs	-40.479%
fft	90	24.815 µs	-42.182%
fft	120	24.746 µs	-42.175%
fft	150	53.449 µs	-42.518%
fft	255	54.293 µs	-42.240%
direct	1	68.245 ns	-26.112%
direct	30	4.4645 µs	-46.017%
direct	60	17.331 µs	-44.944%
direct	90	67.985 µs	-45.110%
direct	120	70.084 µs	-44.915%
direct	150	265.90 µs	-47.659%
direct	255	268.17 µs	-44.727%

[divergentdave]

FWIW this comment is out of date (had me confused for a bit). We changed all field encodings to be little-endian in #495, following a change in draft-irtf-cfrg-vdaf-05. Thus, Prio v2 and VDAF all use the same encoding.

[armfazh]

Closing in favor of #1099