Decide on a consistent semantics for float eq/ord

Currently, AlgebraicValue::F32/F64 uses decorum::Total<f32> and Total<f64> as a way to get Ord and PartialEq implementations for floats. The Ord impl eventually uses decorum::FloatOrd and decorum::FloatEq. For FloatEq, this eventually results in calling to_canonical_bits, which is implemented here. As explained in the FloatOrd and FloatEq docs, one of their properties is that they treat all NaNs as equal. These semantics for floats are used for AlgebraicValue, eq_bsatn, and in eq_to_pv.

Meanwhile, the spacetimedb_table::eq implementation compares bytes directly, which is ostensibly compatible with the {f32, f64}::total_cmp methods, which behave according to the totalOrd predicate defined by IEEE 754 FP standard, and unlike the behavior of FloatOrd.

We thus have an inconsistency in our implementation and should pick one.
For performance, it seems like total_cmp would be the winner, as it allows us to keep comparing bytes directly for VLOs.

Start by comparing AlgebraicValue::F32/F64 and the eq_bsatn and eq_to_pv paths with spacetimedb_table::eq. Read the referenced decorum FloatOrd, FloatEq, and total_cmp semantics, including the byte comparisons used for VLOs. Done means choosing one float equality and ordering policy and making the affected implementations consistent with it.