fix: resolve double LIMIT bug in SQL pagination wrapping

[midasism]

Summary

• Fix a bug where extract_order_by/remove_order_by used naive rfind("ORDER BY") that captured trailing LIMIT/OFFSET clauses, producing double LIMIT syntax errors when the pagination layer appended its own LIMIT.
• Replace per-driver helpers with a shared, quote-and-bracket-aware SQL tokenizer (extract_trailing_clauses + build_paginated_query) in drivers/common.rs that correctly separates ORDER BY, LIMIT, and OFFSET from the base query.
• MCP run_query now accepts an optional limit parameter and respects user-specified LIMIT clauses (takes min(user_limit, page_size)).

Bug details

When a user (or AI agent via MCP) ran a query like:

SELECT * FROM tapp_appointment_message_event_limit ORDER BY id LIMIT 10

The old code produced invalid SQL:

SELECT * FROM (SELECT * FROM tapp_appointment_message_event_limit) as data_wrapper ORDER BY id LIMIT 10 LIMIT 101 OFFSET 0

The double LIMIT 10 LIMIT 101 is a syntax error on all database engines.

Changes

File	Change
src-tauri/src/drivers/common.rs	Add shared tokenize_sql, extract_trailing_clauses, build_paginated_query + 18 unit tests
src-tauri/src/drivers/postgres/mod.rs	Remove old helpers, use shared functions
src-tauri/src/drivers/mysql/mod.rs	Same
src-tauri/src/drivers/sqlite/mod.rs	Same
src-tauri/src/mcp/mod.rs	Add optional limit param to run_query, use configurable max_rows

Test plan

• 29 unit tests pass (cargo test --lib drivers::common::tests) covering:
  • Basic queries without trailing clauses
  • ORDER BY only / LIMIT only / both / all three
  • Table names containing "limit" substring (e.g. tapp_appointment_message_event_limit)
  • Quoted identifiers and string literals with SQL keywords
  • Subqueries with inner ORDER BY
  • Multi-column ORDER BY
  • Trailing semicolons
  • build_paginated_query with user limit smaller/larger than page_size
  • Pagination page 2+ offset calculation
  • SQLite (no alias) vs PG/MySQL (with alias) variants
• Full cargo test --lib passes (115/117 pass; 2 pre-existing failures in updater::tests unrelated to this change)
• cargo build succeeds with no warnings

[debba]

Thanks for PR, I will check it ASAP

[debba]

i will merge it, thanks