Optimize MSM for small scalars

[mratsim]

Modern polynomial IOPs (Interactive Oracle Proofs) try to commit to only small values.

This is especially important for proving hash functions implemented via a bit-based approach (for xor/shift/rotate/...) vs lookup based approach. In that case, many scalars will be just 1.

One such optimization is to not double if the miniMSM coefs are all 0:

constantine/constantine/math/elliptic/ec_multi_scalar_mul.nim

Lines 237 to 254 in dcc9310

	func miniMSM[bits: static int, EC, ECaff](
	r: var EC,
	buckets: ptr UncheckedArray[EC],
	bitIndex: int, miniMsmKind: static MiniMsmKind, c: static int,
	coefs: ptr UncheckedArray[BigInt[bits]], points: ptr UncheckedArray[ECaff], N: int) {.meter.} =
	## Apply a mini-Multi-Scalar-Multiplication on [bitIndex, bitIndex+window)
	## slice of all (coef, point) pairs
	var windowSum{.noInit.}: typeof(r)
	windowSum.bucketAccumReduce(
	buckets, bitIndex, miniMsmKind, c,
	coefs, points, N)
	# 3. Mini-MSM on the slice [bitIndex, bitIndex+window)
	r ~+= windowSum
	when miniMsmKind != kBottomWindow:
	for _ in 0 ..< c:
	r.double()

The bucketAccumReduce function for example might probably be modified to return if there is at-least a non-zero coefficient found so we do a single pass over the data.