Floating-point interval approximation proofs

This entrypoint collects the IEEE32Exec interval-semantics development used by the floating-point universal-approximation development. The files underneath separate the work into:

• interval-domain semantics for executable binary32 networks;
• exact interval-image statements for rounded targets; and
• the constant-target base theorem.

The design keeps the trusted/executable float representation visible while proving the interval claims in Lean over the concrete IEEE32Exec semantics.

Reference:

• Hwang, Lee, Park, Park, and Saad, "Floating-Point Neural Networks Are Provably Robust Universal Approximators", arXiv:2506.16065.