Enhance `Codec`'s decoding options

## Problem
Currently, `Decode` allows decoding only the entire axis. It requires a 2D bytes slice as input and reconstructs nil shares via non-nil ones. However, we now need an additional use case where we want to reconstruct only particular nil shares and not all of them. This optimization is a part of [lazy-extend](https://github.com/celestiaorg/rsmt2d/issues/266) effort and avoids unnecessary share reconstruction for `Codec`

## Solutions
* change the `Decode` slice share inputs to have three states: built, nil, and empty(zero-length). The new `empty` share slices will signal to `Codec` that they need to be reconstructed by the provided `built`, while `nil` shares will be left untouched.
* Less explicit and needs to be properly documented
* add a new `DecodeSome` method that takes the same shares slice as input + an additional slice of bools with the same line as the shares slice, following [klauspost's ReconstructSome](https://github.com/klauspost/reedsolomon/blob/674f18bb1f8189237afb301da140a899a4a85756/reedsolomon.go#L78) method.
* Allocates additional slice
* Two code paths for lazy and non-lazy decoding that can be single.

# Refs

Blocked on https://github.com/klauspost/reedsolomon/issues/264

Start by reading the Decode behavior described in this issue and the linked lazy-extend issue #266. Compare the two proposed approaches, then review the referenced klauspost ReconstructSome implementation and issue #264. Done means Codec can reconstruct only the requested missing shares while leaving other nil shares untouched, with the chosen behavior documented.

Written by the indexing model from the issue text.