Floating point comparisons should not be tested with the validator

[ml-evs]

During validation, we currently test that queries for e.g. _custom_float_field=3.5 returns at least one matching document to _custom_float_field>=3.5 and other inclusive operators. The OPTIMADE specification is quite flexible on floating point comparisons, so during validation we should be too:

While the filtering language supports tests for equality between properties of floating point type and decimal numbers given in the filter string, such comparisons come with the usual caveats for testing for equality of floating point numbers. Normally, a client cannot rely on that a floating point number stored in a database takes on a representation that exactly matches the one obtained for a number given in the filtering string as a decimal number or as an integer. However, testing for equality to zero MUST be supported.

This is a big problem when e.g. testing queries on elements_ratios with filters like elements_ratios HAS 0.299999999999999988897769753748434595763683319091796875, or whatever.