fix(cryptofuzz): expose all cryptofuzz tested primitives in lowlevel_*

[mratsim]

Constantine is continuously fuzzed on Google OSS-Fuzz through CryptoFuzz.

The test harness is https://github.com/guidovranken/cryptofuzz/blob/042cac0727b99a39dfecfd61d994f5972e2e7e3d/modules/constantine/constantine_harness.nim and has been broken by refactoring preceding the v0.1.0 release:

• Elliptic Curve: ⚠️ breaking❗ rename isInf/setinf to isNeutral/setNeutral and rename generator #399
• Named algebra (part 1) #400
• Named algebras 2 (⚠️ breaking internals❗) #402

This PR:

• exposes everything cryptofuzz tests in the lowlevel_*.nim files so that it doesn't depend on internal folder structure.
• remove the new CastSizes warning when raw casting between integer of different size for allocation.
• exposes new pow/pow_vartime by a finite field and out-of-place using ^ and ~^

It also fails to enable creating a generic version of https://github.com/guidovranken/cryptofuzz/blob/042cac0727b99a39dfecfd61d994f5972e2e7e3d/modules/constantine/constantine_harness.nim#L67-L89

proc loadScalar_BN254_Snarks(
       dst: var matchingOrderBigInt(BN254_Snarks),
       src: openarray[byte]) =
    const maxBits = 8 * roundNextMultipleOf(BN254_Snarks.getCurveOrderBitwidth(), 8)
    var tmp{.noinit.}: BigInt[maxBits]
    tmp.unmarshal(src, bigEndian)
    dst.reduce(tmp, BN254_Snarks.getCurveOrder())

proc loadScalar_BLS12_381(
       dst: var matchingOrderBigInt(BLS12_381),
       src: openarray[byte]) =
    const maxBits = 8 * roundNextMultipleOf(BLS12_381.getCurveOrderBitwidth(), 8)
    var tmp{.noinit.}: BigInt[maxBits]
    tmp.unmarshal(src, bigEndian)
    dst.reduce(tmp, BLS12_381.getCurveOrder())

proc loadScalar_BLS12_377(
       dst: var matchingOrderBigInt(BLS12_377),
       src: openarray[byte]) =
    const maxBits = 8 * roundNextMultipleOf(BLS12_377.getCurveOrderBitwidth(), 8)
    var tmp{.noinit.}: BigInt[maxBits]
    tmp.unmarshal(src, bigEndian)
    dst.reduce(tmp, BLS12_377.getCurveOrder())

When replacing by a generic:
matchingOrderBigInt(curve) and curve.getCurveOrder() fail to match types.

Changing to curve.getBigInt(kScalarField) doesn't work either, and changing the internal impl of getBigInt to use typeof(CurveOrder) triggers a compiler crash