fix(raft): repair snapshot compaction

[windsornguyen]

Pull Request

Summary

What: Repair snapshot compaction metadata, suffix retention, snapshot responses, and post-snapshot AppendEntries boundaries.

Why: Raft snapshot installation is not just "clear the log"; the snapshot carries the configuration at its boundary, and a snapshot that describes a log prefix must preserve the valid suffix.

Lines added: +221

Thesis Cross-Reference

• Snapshot metadata includes configuration:
  compaction/memsnapshot.tex#L48-L60
  Key phrase: "latest configuration at that point".
  Context: InstallSnapshot must carry last included index, last included term, and the configuration at that index.

• Snapshot prefix retention:
  compaction/memsnapshot.tex#L69-L78
  Key phrase: "must be retained".
  Context: conflicting/new snapshots supersede the log, but snapshots describing a prefix delete only covered entries.

• Compaction invalidates contiguous-log assumptions:
  compaction/memsnapshot.tex#L244-L260
  Key phrase: "might have been discarded".
  Context: every log lookup around prev_log_index must handle compacted prefixes.

• AppendEntries consistency boundary:
  basicraft/consensus.tex#L320-L334
  Key phrase: "same index and term".
  Context: the snapshot boundary term is needed so followers can satisfy AppendEntries matching at lastIncludedIndex.

Test Plan

• cargo test -p cloud9-raft

Tests Added

• Snapshot boundary returns snapshot_term for AppendEntries matching.
• Snapshot truncation preserves the latest configuration.
• Installing a matching-prefix snapshot retains the suffix.
• Installing a conflicting snapshot discards the suffix.
• Followers reject snapshots that would discard committed suffix state.
• Leaders advance follower progress from InstallSnapshotResponse.

Notes for Reviewers

The important invariant is that compaction turns snapshot_index into the synthetic log entry immediately before first_index. Treating its term as unavailable breaks replication after compaction.