fix(joins): single-equality RIGHT JOIN layout + operand routing (P63, P64)

[TimelordUK]

Summary

• P63: a single-equality RIGHT JOIN took the hash path as a LEFT join with the tables swapped and kept that swapped layout, so a.*/b.* were exchanged on every row (P8's fault, on the hash path). New hash_join_right, sharing right_join_result_table (factored out of nested_loop_join_right_multi) so both RIGHT builders use one layout.
• P64 (found by P63's regression test): the simple-column resolver routed operands by where they are written, not by alias, so b.id = a.partner_id ran as a.id = b.partner_id whenever both tables hold both columns (self-joins). This hit INNER, LEFT and RIGHT. Operands are now routed with P7's operand_uses_right and swapped, with the operator reversed, before resolution.
• Removed unused column-name params on the inner/left hash builders.

Verification

• Parity 197 → 202 AGREE / 233. The diff against the saved baseline moves exactly the 2 P63 cases and the 3 new P64 cases; the P64 cases were pinned as DIFFER on main first. --check holds.
• New Rust tests in tests/join_operand_order_tests.rs over an asymmetric edges table: both operand orders, plus hash path vs nested loop.
• cargo test, Python tests (572 passed) and FORMAL examples are green. expander_rewriters smoke also fails on main (unrelated).

🤖 Generated with Claude Code