[devin-ai-integration]

Summary

E2E tests (and, in follow-up commits on this PR, fixes) for ways Ruby applications commonly use the generated code that were not covered yet.

New scenarios (PostgreSQL / MySQL / SQLite where applicable):

• forked workers (Puma / Sidekiq / Resque) after the parent ran queries, pooled and single connections
• a raw connection shared with another library (ActiveRecord raw_connection, Sequel): PG type maps / field_name_type, mysql2 query_options and Mysql2::Client.default_query_options, SQLite results_as_hash
• QuerySql.connect(DATABASE_URL) on PostgreSQL
• #transaction inside a transaction opened by another library
• schema-per-tenant search_path switching with prepared statements
• primary + read-replica query objects
• memoized frozen query objects, subclasses / prepended wrappers forwarding ... / **, SimpleDelegator
• rows through Marshal (Rails.cache) and JSON

Failing on main (a44f282):

• pg test_a_forked_child_leaves_the_parent_connections_usable (child's connection_pool after_fork closes the inherited connections and terminates the parent's sessions)
• pg test_a_connection_shared_with_another_library_keeps_its_settings (generated code overwrites the connection's type maps; field_name_type = :symbol raises UndefinedEncoder)
• pg test_connect_accepts_a_connection_url (TypeError)
• mysql2 test_a_forked_child_leaves_the_parent_connections_usable (same as pg)
• mysql2 test_a_client_with_other_query_options_still_returns_typed_rows (cast: false returns Strings, cast_booleans: true returns true/false for Integer columns)

Fixes (generated code, lib/sqlc_gen_ruby/drivers/{pg,mysql2}.rb):

• pg: generated queries pass their own type maps (exec_prepared / exec_params with client.query_types, results get client.result_types and field_name_type = :string), so the connection's type_map_for_results / _for_queries / field_name_type are never changed. QuerySql.connect also takes a URL string.
• pg / mysql2: ConnectionState remembers the pid that opened the connection; when connection_pool's after_fork closes it in a child, its socket is first pointed at /dev/null, so the parent's session survives. A pg connection used directly in a child reconnects; a direct mysql2 client raises Mysql2::Error::ConnectionError.
• pg / mysql2: a Process._fork hook releases, in the child, every inherited connection the generated code has used (ObjectSpace.each_object(ConnectionState)), also ones another thread had checked out at fork time and ones the child never uses, so the child's exit does not end the parent's sessions. Closing an already closed pg connection is a no-op (connection_pool's after_fork closes those too).
• mysql2: rows are read with as: :array, cast: true, cast_booleans: false whatever the client's / default query options.
• Documented (README, Limitations): MySQL transaction inside a raw BEGIN commits it; SQLite children must reopen the database. docs/Coverage.md lists every new test.

Checklist

• examples/ is regenerated (rake sqlc_generate_process) and committed
• Tests added or updated (rake test, and rake end2end_* when the generated code changes)
• The generated code stays RuboCop-clean (rake rubocop) and Steep-clean (rake steep)
• Labelled (bug, enhancement, documentation, or breaking) so the release notes are grouped correctly

Requested by: @doron050

[devin-ai-integration]

I'll fix CI failures and address comments from users with write access. I'll skip comments containing "(aside)".

• Disable automatic comment, CI, and merge conflict monitoring