adds tool test showing problem with Infinity handling

[bernt-matthias]

Follow up on https://github.com/galaxyproject/galaxy/pull/9425 ...

Problem seems to be that delta_frac is encoded to json at some point where Inf is encoded by '__Infinity__' but decoding is not doing the back translation.

So far I could not find another attribute value that is affected.

This special treatment of Inf, -Inf and NaN in Galaxy has been introduced here: https://github.com/galaxyproject/galaxy/commit/8f72dfe8ee1a3e4394976ef272057961030891c0

Possibilities:

• use python's default handling of encoding Inf as the string 'Infinity', .... The downside would be that all occurrences of 'Infinity' would be affected. So '__Infinity__' seems to add a a bit more safety. But it would be faster since swap_inf_nan becomes obsolete.
• implement a backtranslation function like so: https://github.com/galaxyproject/galaxy/commit/8f72dfe8ee1a3e4394976ef272057961030891c0 which has the problem that it adds run time (like swap_inf_nan which does the translation during encoding).
• subclass the encoders / decoders of the python json module and overwrite the handling of Inf, ... But since this seems quite hardcoded (https://github.com/python/cpython/blob/b0be6b3b94fbdf31b796adc19dc86a04a52b03e1/Lib/json/encoder.py#L231) I don't see an obvious way to do this future proof.

[mvdbeek]

I get the vague feeling that attaching pydantic models to parameters and attributes could help us here, since then we don't blindly need to guess if something is NaN / Infinity etc. Let me see how much work this would be.