Support nullable matrix IDs

[luisleo526]

What changed

• add a nullable validity state to float, generic, and bool matrix IDs
• distinguish a null matrix ID from a valid empty matrix
• make matrix operations fail deterministically on a null receiver or operand before shape/index checks
• preserve valid-empty behavior for existing mathematical failure results
• add focused runtime coverage for null, empty, copy/move, result construction, and operation guards

Why

Pine permits na for matrix IDs. PineForge previously represented a default matrix as an empty matrix, so it could neither implement na(matrix_id) nor distinguish null-ID lifecycle errors from valid empty-matrix behavior. That runtime gap combined with a separate codegen target-typing gap.

Impact

Generated strategies can now carry a real nullable matrix ID once the matching codegen follow-up lands. This is an internal C++ layout change, so consumers must cleanly rebuild and must not mix old objects with the new library.

This PR deliberately does not change matrix assignment alias semantics, nullable arrays, matrix-returning UDF inference, or UDT identity behavior.

Validation

• clean rebuild and CTest: 115/115 passed
• engine corpus: 252/252 executed; 247 Excellent / 4 Strong / 1 documented anomaly
• exhaustive cross-repository N=3 matrix: 8/8, early stop disabled, publication gate passed
• forced-no-cache scraper gate: 416/416 accounted, zero compile/driver/unresolved timeout, 0 DOWN
• independent final review: no in-scope blocker
• production diff SHA-256: 89e43bf85aab0842ef731419e354a0fee6244d4b1c98d4d10b3999276c5b7254