feat: Handling Panics in HashNode Method

[staheri14]

Problem

Identified as part of this EPIC: celestiaorg/celestia-app#1296.
The HashNode method of the NMT may potentially cause a panic if it receives invalid or malformed inputs (as mentioned by @evan-forbes in this comment #102 (comment)). For example, if the left or right children do not conform to the namespace hash format, meaning their length is less than the namespace ID size, or if the relation between the left and right children namespace range is invalid. If an invalid node is sent to a light node, it can result in a crash.

It is important to note that returning an error from this method is not possible as it would violate the interface it is intended to fulfill i.e., NodeHasher and Hash.

Acceptance Criteria

This issue is to implement proper error handling mechanism for HashNode.

[staheri14]

Reopening this issue as it is not completely resolved.
As mentioned in the last relevant PR i.e., #113, to avoid situations in which light clients may panic, we need to integrate the node validity checks into the proof verification and construction and return error for invalid inputs. Also, as per conversation with @evan-forbes, we also decided to change the signature of the HashNode and HashLeaf to return errors for invalid inputs instead of panic. All these require some refactoring. Once done, we can close this issue.