fix(DM01-6184): filter migrations by file number + skip comment-only files

[Jiachen0715]

Summary

Builds on #636 (fix(DM01-6184): filter migrations by file number instead of list index) and adds a second fix that is needed to fully unblock the infrabox-db migration Job.

Two independent bugs, one root symptom

test-new databases starting below schema_version=46 currently see the migration Job crash with:

psycopg2.ProgrammingError: can't execute an empty query

There are actually two bugs at play, so #636 alone is not enough. This PR keeps #636's fix and adds the missing one.

Bug 1 — get_files used list index instead of file number ✅ fixed by #636

Historically 00046.sql did not exist (the sequence went 45 → 47), and files = files[current_schema_version:] used list index to slice pending migrations. Any gap in the numbering desynchronised the index from the file number, silently skipping later migrations. #636 replaces the slice with a file-number filter — kept as-is in this PR.

Bug 2 — apply_migration treats comment-only files as executable ← this PR adds

Even with #636's filter selecting the correct files, once 00046.sql (an intentionally empty placeholder that only contains -- ... lines) is chosen, the runner still fails:

sql = sql_file.read().strip()   # keeps `--` comment lines
if sql:                          # non-empty string of comments → guard passes
    cur.execute(sql)             # psycopg2 rejects: "can't execute an empty query"

The result: the migration Job crashes before ever reaching 00047.sql (which creates mcp_token, mcp_access_log, and adds build.source) or 00048.sql (secret_read_token). Downstream, MCP-related endpoints return 500 because the tables don't exist.

Fix in this PR

Add a small _has_executable_sql() helper that returns True only if the file has at least one non-blank line that isn't a -- comment. apply_migration uses it to decide whether to call cur.execute():

if _has_executable_sql(sql):
    cur.execute(sql)
else:
    logger.info("Skipping execute for %s (no non-comment SQL statements)", filename)

schema_version is still bumped afterwards, so the placeholder still serves its purpose (closing a numbering gap).

Scope

• Kept: the get_files fix from fix: filter migrations by file number instead of list index #636 (unchanged).
• Added: _has_executable_sql helper + guard swap in apply_migration.
• Only src/db/migrate.py changes. No migration file, no DB schema, no API/UI impact.

Behaviour after this PR

File type	Before	After
Regular .sql with statements	executed	executed (unchanged)
.sql with only -- comments	crashes with "empty query"	logs "Skipping execute", schema_version advances
.py migration	import + migrate(conn)	unchanged

Test plan

• Reproduce the failure locally: point migrate.py at a DB at schema_version=45, run against a migrations folder containing the current 00046.sql (comment-only) + 00047.sql. Confirm the pre-fix code crashes and the post-fix code logs Skipping execute for ...00046.sql and completes.
• Deploy to a real environment (e.g. test-new) that is currently stuck at schema_version=45. Expect the infrabox-db Job to succeed and MCP endpoints (GET/POST /api/v1/mcp/tokens) to return 200 instead of 500.

Notes

• _has_executable_sql() is a lightweight line-based heuristic. It only recognises -- line comments; it does not parse SQL and does not understand /* block */ comments. That's fine for the current use case (the placeholder file uses -- comments), but if we ever want to be stricter we can swap in a real tokenizer later.
• This PR is complementary to the effort in SAP/InfraBox#637 (unrelated feature branch) which also had to work around this same crash. Once this PR merges, that workaround (adding SELECT 1; inside 00046.sql) becomes unnecessary and can be reverted.

[Jiachen0715]

Superseded — pushed the additional commit directly to the PR #636 branch (DM01-6184) instead.