fix(MSM): properly handle edge condition in parallel MSM when bits is exactly divided by c

[mratsim]

The parallel MSM skips the top window if the number of bits is exactly divided by MSM parameter c.

This is easy to reproduce with 3 points and by removing the endomorphism acceleration (which multiply the number of points by 2x or 4x) here:

constantine/constantine/math/elliptic/ec_multi_scalar_mul_parallel.nim

Lines 539 to 557 in 7c3d76d

	proc multiScalarMul_dispatch_vartime_parallel[bits: static int, F, G](
	tp: Threadpool,
	r: ptr (EC_ShortW_Jac[F, G] or EC_ShortW_Prj[F, G]),
	coefs: ptr UncheckedArray[BigInt[bits]],
	points: ptr UncheckedArray[EC_ShortW_Aff[F, G]], N: int) =
	## Multiscalar multiplication:
	## r <- [a₀]P₀ + [a₁]P₁ + ... + [aₙ]Pₙ
	let c = bestBucketBitSize(N, bits, useSignedBuckets = true, useManualTuning = true)
	# Given that bits and N change after applying an endomorphism,
	# we are able to use a bigger `c`
	# but it has no significant impact on performance
	case c
	of 2: withEndo(msmImpl_vartime_parallel, tp, r, coefs, points, N, c = 2)
	of 3: withEndo(msmImpl_vartime_parallel, tp, r, coefs, points, N, c = 3)
	of 4: withEndo(msmImpl_vartime_parallel, tp, r, coefs, points, N, c = 4)
	of 5: withEndo(msmImpl_vartime_parallel, tp, r, coefs, points, N, c = 5)
	of 6: withEndo(msmImpl_vartime_parallel, tp, r, coefs, points, N, c = 6)

This might help for #479, cc @Vindaar

[mratsim]

failure go test on Windows

[mratsim]