queuefs: harden queue-table identifier handling beyond sanitizeTableName

[JaySon-Huang]

Summary

queuefs currently relies on sanitizeTableName(queueName) as the main boundary before interpolating queue table names into SQL statements. On upstream/master, that helper only rewrites /, -, and . before the resulting identifier is used in CREATE TABLE, SELECT, UPDATE, DELETE, and DROP TABLE statements.

Current upstream code

On upstream/master:

• agfs-server/pkg/plugins/queuefs/db_backend.go:232 defines sanitizeTableName(queueName string)
• it currently rewrites only /, -, and . and prefixes queuefs_queue_
• the resulting table name is then interpolated directly into SQL statements in the SQL-backed queue backend

Why this needs follow-up

This is fragile as a long-term identifier strategy:

• it depends on partial character replacement rather than a strict identifier policy
• it does not centralize backend-specific identifier quoting/escaping
• different SQL dialects have different rules for unquoted identifiers
• newer SQL backend refactors reuse the same table-name strategy across SQLite, TiDB/MySQL, and PostgreSQL, so the impact surface grows

Even if most queue names are path-like today, using partial sanitization as the safety boundary for SQL identifiers is brittle and should be tightened.

Suggested direction

A follow-up fix should move to one of these stricter approaches:

1. define a strict normalized identifier alphabet (for example [A-Za-z0-9_] after normalization) and reject/transform everything else deterministically
2. apply dialect-specific identifier quoting consistently when interpolating table/index identifiers
3. ideally do both: normalize first, then quote when building SQL

The goal is to stop treating sanitizeTableName as sufficient identifier hardening by itself.