fix: guard delete detection against non-integer/large primary keys (#42)

[y-ken]

Summary

Fixes the mysql_replicator: failed to execute query / bad value for range error reported in #42.

Root cause

On the first poll (no previous snapshot yet), delete detection ran:

deleted_ids = [*1...current_ids.max] - current_ids

This builds an integer Range from the primary-key values, which:

• raises bad value for range when the primary_key is non-integer (e.g. a UUID or string id) — exactly mysql_replicator: failed to execute query.error #42;
• for large ids, allocates a huge array ([*1...10_000_000]);
• emits phantom delete events for ids that never existed (every integer missing between 1 and max).

Fix

Extract the logic into detect_deleted_ids(previous_ids, current_ids):

def detect_deleted_ids(previous_ids, current_ids)
  return [] if previous_ids.empty?   # first poll: baseline only
  return [] if current_ids.empty?    # don't mass-delete on an empty result
  previous_ids - current_ids
end

The first poll now only establishes a baseline (reports nothing deleted); deletions are detected on subsequent polls by diffing against the previous snapshot. All other behavior is unchanged.

Tests

Added unit regression tests (no DB needed) covering:

• first poll with string keys → no bad value for range;
• first poll with sparse integer ids → no phantom deletes;
• normal delete diffed against the previous snapshot;
• emptied result set → no mass delete.

Verified locally on Ruby 3.4 (6 tests, 11 assertions, 0 failures).

Closes #42

🤖 Generated with Claude Code