fix(security): add SQL injection prevention to ExplainPlanTool

[RajatRaiD11]

Summary

The explain_query tool is annotated with readOnlyHint=True, but in UNRESTRICTED mode (default), user-supplied SQL is concatenated via f-string into EXPLAIN queries without validation. This allows:

1. Multi-statement SQL injection — e.g., SELECT 1; DROP TABLE users passes through f"EXPLAIN (FORMAT JSON) {sql_input}" and reaches cursor.execute() which supports multi-statement execution via cursor.nextset()
2. 2. EXPLAIN ANALYZE DML execution — EXPLAIN ANALYZE DELETE FROM users actually executes the DELETE, but the tool doesn't restrict this

What this PR does

Adds _validate_explain_input() static method using pglast (already a project dependency) to:

• Parse input SQL and reject multi-statement injection attempts
• • Restrict EXPLAIN ANALYZE to SELECT-only (since it executes the query)
• • Validate syntax before f-string concatenation
    Integrates validation into explain(), explain_with_hypothetical_indexes(), and generate_explain_plan_with_hypothetical_indexes().

What the codebase already gets right

• SafeSqlDriver provides excellent pglast-based validation in RESTRICTED mode
• • psycopg.sql.Literal() is used correctly for IndexDefinition (not exploitable)
• • Read-only transaction enforcement in RESTRICTED mode is solid
    This fix extends the same validation approach to UNRESTRICTED mode for the EXPLAIN tool specifically.

Files changed

• src/postgres_mcp/explain/explain_plan.py — adds _validate_explain_input() and integrates it (57 additions, 0 deletions)
• • tests/unit/explain/test_explain_input_validation.py — 22 test cases (17 unit + 5 integration)

Test coverage

• Valid queries (SELECT, JOIN, CTE, subquery, INSERT/UPDATE/DELETE without ANALYZE, bind params)
• • Multi-statement injection rejection (semicolons, comments, COPY exfiltration)
• • Empty/invalid input handling
• • EXPLAIN ANALYZE DML restrictions (blocks DELETE, UPDATE, INSERT; allows SELECT)
• • Integration tests verifying explain(), explain_analyze(), and explain_with_hypothetical_indexes() block injection and never send malicious SQL to the database