perf(results): avoid double-copy compute for row→string conversion

[ZhuchkaTriplesix]

Parent

Part of #414

Problem

Postgres / MySQL / SQLite workspaces build List<List<Object?>>, then compute(...) when length > 500. Cross-isolate copy of the whole matrix often dominates toString work → peak ≈ 2× row matrix for large caps (up to 100k).

Evidence

• lib/features/postgresql/postgres_sql_workspace.dart (~353–365)
• lib/features/mysql/mysql_sql_workspace.dart (~195–213)
• lib/features/sqlite/sqlite_sql_workspace.dart (~172–179)
• lib/features/postgresql/postgres_table_view.dart (~190–194)

Acceptance

• Large results convert without shipping a second full matrix copy when avoidable
• UI stays responsive; document strategy (stream convert / single background isolate / yield loop)
• Benchmark or comment with measured approach for 10k+ rows

Suggested fix

Convert while streaming on one background isolate; or in-place convert with yields for mid-size results.

[ZhuchkaTriplesix]

Shipped on dev via #437 (Closes did not auto-close because PRs targeted dev).