frontend/testfloat_expr Design

The package parses TestFloat function metadata and fixed-width hexadecimal vectors, decodes operands through BinaryInterchange, constructs a BinaryContext, executes add/sub/mul/div/sqrt, and compares encoded result and all five IEEE flags. Sharding is stable by row index.

Accepted formats are binary16/32/64/128 and accepted operation, rounding, and tininess modes are exactly those parsed by TestFloatSpec. The package does not generate vectors, invoke SoftFloat, read files, or extend coverage beyond that matrix.