fix(core): tolerate missing columns on legacy v1 session table

[BambinoSK]

Issue for this PR

Closes #54009

Type of change

• Bug fix
• New feature
• Refactor / code improvement
• Documentation

What does this PR do?

The V1→V2 data migration fails on every startup with SQLiteError: no such column: path when the legacy session table predates newer columns. The per-session transaction (including the cursor checkpoint) rolls back, the migration.v1-v2 marker stays stuck in the sessions phase, and the import is retried and fails again on every server start.

Root cause: v1-migration.bun.ts copies each legacy row into session_v2 with a fixed INSERT ... SELECT column list that unconditionally references path, metadata, cost, tokens_*, agent, and model on the source table. V1 databases last migrated before those columns were added do not have them, so statement preparation fails. The opencode-next.db import already handles this class of problem through selectNextRows (pragma-based column projection); the legacy copy had no equivalent.

The fix reads the legacy session table's columns with pragma_table_info before the copy loop and builds the INSERT ... SELECT from only the columns the table actually has. Missing columns are omitted from the insert so session_v2 defaults fill them (cost/tokens_* default to 0, nullable columns to NULL). permission was previously inserted explicitly as NULL and now falls back to its default, preserving the intentional clearing of V1 permission state.

How did you verify your code works?

• Added a regression test (imports legacy V1 sessions from tables missing newer columns) that creates a legacy session table without path, metadata, cost, tokens_*, agent, and model — the shape observed in the issue — with one session, message, and part row. It verifies the migration completes, session_v2 receives defaults for the missing columns with agent/model derived from message data, the projected session_message row is written, and the marker reaches {"phase":"completed"}.
• Confirmed the new test fails against the unfixed source and passes with the fix.
• Full packages/core/test/v1-migration.test.ts run: 27 pass, 0 fail.
• bun typecheck in packages/core passes.

Note: the local pre-push bun run check could not run on Windows because oxlint fails to load its TypeScript plugin via Node ESM (Unknown file extension ".ts"); that failure is environmental and unrelated to this change.

Screenshots / recordings

Not a UI change.

Checklist

• I have tested my changes locally
• I have not included unrelated changes in this PR

[github-actions]

Thanks for your contribution!

This PR doesn't have a linked issue. All PRs must reference an existing issue.

Please:

1. Open an issue describing the bug/feature (if one doesn't exist)
2. Add Fixes #<number> or Closes #<number> to this PR description

See CONTRIBUTING.md for details.

[github-actions]

The following comment was made by an LLM, it may be inaccurate:

[opencode-agent]

This looks good. I ran the reproduction script from #54009 against the CLI from source. On the base branch, every start logs no such column: path and the marker stays at sessions. With this change the migration completes, the legacy session is copied into session_v2 with defaults for the missing columns, and the marker reaches completed. The v1-migration tests (27 pass) and the packages/core typecheck both pass.

I also checked that session_v2.permission is nullable with no default, so leaving it out still clears V1 permission state as before. Column names come from a fixed list, are filtered against pragma_table_info and are quoted with sql.identifier. I have no concerns.

[github-actions]

Thanks for updating your PR! It now meets our contributing guidelines. 👍