Floats entrypoint

This is the “all floating-point semantics in one import” entrypoint:

• proof-oriented real-valued models (FP32, NeuralFloat),
• the executable bit-level model (IEEEExec),
• interval/enclosure utilities (Interval),
• and the external Arb oracle integration (Arb),
• and the shared error-bound vocabulary used across the library.

This module is the chapter boundary for the floating-point surface. It imports the focused NN.Floats.* subsystems directly so downstream users have one stable entrypoint.