fix: fft domain precompute thread safety

[gbotrel]

Missed that in my review of #45 .

The pattern

BitReverse(domain.CosetTableInv[coset-1])
// ...  scale by coset
BitReverse(domain.CosetTableInv[coset-1])

is not thread safe, and quite wasteful.

Replaced by dynamically computing the correct index in the coset table:

parallel.Execute(len(a), func(start, end int) {
	for i := start; i < end; i++ {
		irev := bits.Reverse64(uint64(i)) >> nn
		a[i].Mul(&a[i], &domain.CosetTable[coset-1][int(irev)])
	}
})

Benchmark:

BenchmarkFFT/fft_2**19bits_(cosets_without_precomputations)-8     63984013      43923680      -31.35%

Also did few style adjustment with less if -> if -> else -> if statements.

[CLAassistant]

All committers have signed the CLA.