Implement deferred schema cleanup when applying multiple migrations

[blakewatters]

Summary

When pgroll migrate applies multiple outstanding migrations, it breaks zero-downtime guarantees by dropping the previous version schema during each intermediate Complete() call. If an application is still connected via that schema, its queries break immediately.

Example: App runs on public_v1. Migrations v2, v3, v4 are pending:

1. Start v2 → creates public_v2
2. Complete v2 → drops public_v1 (app breaks here!)
3. Start v3, Complete v3 → drops public_v2
4. Start v4 → creates public_v4

This PR implements deferred schema cleanup so that intermediate Complete() calls skip the DROP SCHEMA step, and a batch cleanup runs after all migrations finish — dropping only the intermediate schemas while preserving the original (app-connected) and latest versions.

Changes

• pkg/roll/execute.go: Added CompleteOption functional option type with WithSkipSchemaDrop() to make Complete() conditionally skip dropping the previous version schema. Added DropVersionSchemasExcept() method to drop all version schemas except a specified keep list.
• cmd/start.go: Threaded CompleteOption through runMigration() so callers can forward options to Complete().
• cmd/migrate.go: Updated batch orchestration to record the original version before the batch, pass WithSkipSchemaDrop() for intermediate migrations, and call DropVersionSchemasExcept() after the batch to clean up intermediates while preserving the original and latest schemas.
• pkg/roll/execute_test.go: Added four integration tests covering skip-drop behavior, full multi-migration batch flow with cleanup, the cleanup method itself, and unchanged single-migration behavior.

Behavior after this change

Scenario	Before	After
Multi-migration batch	Each intermediate Complete() drops the previous schema	Intermediate drops are skipped; cleanup runs after the batch
Single migration start/complete	Drops previous schema	Unchanged — WithSkipSchemaDrop() is never passed
Batch failure midway	Intermediate schemas dropped, app broken	Intermediate schemas left behind (harmless views), cleaned up on next successful migrate

Test plan

• Existing test suite passes (go test ./...)
• TestCompleteWithSkipSchemaDropPreservesPreviousVersionSchema — verifies WithSkipSchemaDrop prevents schema drop
• TestMultiMigrationBatchPreservesOriginalSchema — end-to-end batch flow: original schema persists through intermediates, cleanup drops only intermediates
• TestDropVersionSchemasExceptKeepsSpecifiedSchemas — cleanup method keeps specified schemas and drops the rest
• TestSingleMigrationCompleteStillDropsPreviousSchema — normal single-migration behavior is unchanged
• Manual test: create a schema with a table, apply 3+ migrations via pgroll migrate ./migrations, verify original schema persists until cleanup

🤖 Generated with Claude Code