Snap theta before the RXYXY2Q zero check and document the gate

[ciaranra]

Closes #805. Addresses the zero-check half of #804.

Change

• try_rxyxy2q compared theta to exact Angle64::ZERO before its Clifford check, whereas try_rxx snaps within the Clifford tolerance, so a theta within tolerance of zero combined with a non-Clifford phi was rejected although the gate is the identity. It now snaps first, as try_simplify_rxy1q does. The existing zero-theta test covers a near-zero theta too; restoring the exact comparison makes it fail.
• docs/user-guide/gates.md gains an RXYXY2Q section (definition, matrix, relation to RXX/RYY, Selene RPP correspondence, stabilizer angle grid) and a Quick Reference row; gate-naming-conventions.md and gate-angle-types.md list the gate alongside RXY1Q.

Not in this PR

Accepting theta = pi with phi an odd multiple of pi/4 (a Clifford that Selene's Stim simulator accepts) stays open in #804: try_simplify_rxy1q returns a single named GateType, so that case needs a decomposition return type and an eighth-turn snap grid, which is a design change.

Verification

cargo test -p pecos-simulators --lib clifford_rotation (35 passed), cargo clippy --locked -p pecos-simulators --all-targets -- -D warnings, cargo fmt --check, and pre-commit on the changed files, all clean.

Written directly by Claude Fable 5.1 in Claude Code at the maintainer's request; no Codex arm was used for this one.