[committerpb] Add the abort snapshot transaction status and key encoding

[cendhu]

Type of change

• New feature

Description

• SnapshotState.Status gains ABORTED = 6: terminal, meaning the snapshot will not be
  hashed. Written only by the validator-committer when an abort snapshot transaction
  commits, the rule CHECKPOINTED already follows, and never by the snapshot hasher.
• MALFORMED_SNAPSHOT_INVALID_ABORT_KEY = 122: an abort transaction's read-write key does
  not decode as an abort key.
• REJECTED_SNAPSHOT_ABORT_NO_SUCH_SNAPSHOT = 123: a well-formed abort key naming a block
  that is not the snapshot awaiting its checkpoint, or one whose lifecycle is already
  closed. Both are per-transaction rejections rather than failures, because the submitter
  chose the block number and local state is not in question.
• MALFORMED_SNAPSHOT_NOT_MARKER_ONLY's comment is updated: a _snapshot TX is now valid
  in two shapes, a marker-only snapshot request and an abort carrying exactly one
  read-write.
• SnapshotAbortKey(blockNum) builds the _snapshot write key of an abort snapshot
  transaction: the \xffabort/ prefix followed by the block number, encoded order-preserving
  as a checkpoint key is. The leading raw 0xff byte cannot occur in valid UTF-8 transaction
  IDs, which the sidecar already enforces; no hex-ID assumption is needed.
• IsSnapshotAbortKey inspects the prefix only, so a prefixed key whose block number does
  not decode is rejected as a malformed abort rather than mistaken for a snapshot request.
• BlockNumFromSnapshotAbortKey decodes the block number and rejects trailing bytes rather
  than ignoring them: accepting a prefix would attribute the abort to the wrong snapshot.
• snapshot.pb.go and status.pb.go regenerated via make proto.

Additional details (Optional)

The key encoding lives in this module rather than in fabric-x-committer because the only
legitimate submitter of an abort is an administrator, whose tooling depends on this module.

Related issues

• resolves partly [committerpb] A snapshot that can never be hashed has no terminal state #190

[coveralls]

coverage: 82.041% (+0.04%) from 82.004% — cendhu:snapshot-abort into hyperledger:main

[liran-funaro]

The issue is that we never actually verify this.
If we did, we could decode the TX-ID into binary and reduce the write size to the DB and over the network.
This was flagged as an issue during the evaluation, but I dismissed it because we never enforce "hex-encoded digest".
If we want to rely on it, we must enforce it in the sidecar.

[cendhu]

The hex-ID assumption was not enforced. Changed the abort prefix to \xffabort/, where \xff is the raw byte 0xff. The sidecar already requires transaction IDs to be valid UTF-8, which cannot contain this byte, so the key distinction no longer relies on hex IDs or requires a new TX-ID restriction.

Added regression coverage for abort/foo, a transaction ID matching the old abort-key encoding, Unicode IDs, and malformed binary-prefixed keys. make test passed all 1,614 tests with the race detector; make lint passed.

This updates the abort-key encoding introduced by this PR; callers must use the updated helper.