[AbiCoder] Provide coerceFunc for encode / Writer

[silasdavis]

Is your feature request related to a problem? Please describe.
The coerceFunc provides a useful visitor pattern for transforming how values are returned from a decode. Its utility lies in the fact AbiCoder knows how to walk the tree of type definitions so you can decide how to format a value based on the primitive type in a nested array or tuple.

However there is no corresponding facility on the decode side, but the same functionality would be useful. For example I have code like:

// This is run before doing an `encode`
function burrowToAbi(arg: unknown, type: string): unknown {
  if (/address/.test(type)) {
    return recApply(prefixedHexString, arg as NestedArray<string>);
  } else if (/bytes/.test(type)) {
    return recApply(toBuffer, arg as NestedArray<string>);
  }
  return arg;
}

// This is run after doing a `decode` and could be replaced by `coerceFunc`
function abiToBurrow(arg: unknown, type: string): unknown {
  if (/address/.test(type)) {
    return recApply(unprefixedHexString, arg as NestedArray<string>);
  } else if (/bytes/.test(type)) {
    return recApply(unprefixedHexString, arg as NestedArray<string>);
  } else if (/int/.test(type)) {
    return recApply(numberToBurrow, arg as NestedArray<BigNumber>);
  }
  return arg;
}

These functions are crude though because they don't actually understand compound/complex types. They match using a regular expression then have a best effort attempt at converting recursively but they cannot handle mixed types in tuples.

The latter function could be replaced by the existing coerceFunc, but I cannot replace the first in the same way.

Describe the solution you'd like
For symmetry it would be nice to have new AbiCoder({ coerce?: { encode?: CoerceFunc, decode?: CoerceFunc }), whereby you can supply a coerce function that is run before encoding a value in a similar way to which existing coerce functions runs after decoding a value.

Describe alternatives you've considered
See regex hack above. Also not needing to coerce before encoding, but I am dealing with types from a separate GRPC schema that gives types that are in a difference schema.

Additional context
It would be nice to have access to parsed out elements of a type (for CoerceFunc and more generally). For example bitWidth and length, e.g. for the type bytes16[4] you would have 16 and 4. Then I wouldn't have to dirty my hands with string parsing at all ;-)

See #915 for original motivation for coerceFunc