Add JSON column support for nested object indexing

[y-ken]

Summary

MySQL JSON columns (MySQL 5.7.8+ / 8.x) are returned by the driver as plain
strings, so they reached Elasticsearch as escaped strings instead of nested
objects (e.g. "geometry":"{\"type\":\"Polygon\",...}"). This made them
unusable for indexing.

This PR adds an opt-in json_columns option that parses the listed columns'
values into nested objects before they are emitted, so they are indexed as real
nested JSON.

This implements the request from #18 in a type-safe, opt-in way (instead of the
unconditional string replacement proposed there). It supersedes #18.

Changes

• mysql_replicator (single): new json_columns config_param (comma-separated).
• mysql_replicator_multi: new json_columns column on the settings
  management table (varchar(255)). Existing installs can migrate with:

  ALTER TABLE settings ADD COLUMN `json_columns` varchar(255) DEFAULT NULL AFTER `primary_key`;

• Malformed JSON, non-string values, and unlisted columns are left untouched, so
  the default behavior is unchanged (no side effects).
• Documented in the README that this targets Elasticsearch — leave it empty for
  Solr or destinations that cannot store JSON objects.

Notes / scope

• Only top-level columns are parsed (columns inside nested documents are not).
• Diff detection is unaffected: the single plugin hashes the raw string, the
  multi plugin hashes the parsed value, and MySQL returns JSON in a normalized
  form so both are stable across scans.

Testing

• Unit tests (Ruby 3.4): config parsing + parse_json_columns! (valid
  object/array, malformed JSON, non-string, unlisted, empty). All plugin tests
  pass, no regressions.
• E2E (MySQL 8.4 + Elasticsearch 8.18): both single and multi pipelines
  assert a JSON column is replicated as a nested object
  (_source.profile.city == "Tokyo", _source.profile.tags == ["a","b"])
  alongside the existing insert/update/delete checks.

🤖 Generated with Claude Code