fix: timedelta parsing for int and floats

[tcleonard]

Version 11.2.0 introduced a regression when it comes to providing timedelta as floats or integers.

toto = env.timedelta("TOTO", default=30)   # does not works
toto = env.timedelta("TOTO", default=30.0)   # does not works
toto = env.timedelta("TOTO", default="30")   # works
toto = env.timedelta("TOTO", default="30.")   # does not works

This MR aims at making those 4 cases valid

[ddelange]

looks like a lot of logic was added just to bypass the parent class' validation (this line won't be hit anymore).

• a float as default value was previously slipping through the cracks (silent downcast to int before #366, a bug not a feature imo)
• a float-in-string as default value didn't slip through the cracks (expected behaviour to reject)

is this PR something that should rather fold into the bigger topic of #270 and/or #297 instead?

[ddelange]

opened a minimal version to avoid breaking change: #369

[tcleonard]

(this line won't be hit anymore).

It will if you provide something that is of other type than str, float, int, bool, timedelta.
So this is more a last option catch-all.

We could definitely do the more minimal change to get to the previously working (but unconsistent/unlogical behavior)... but I'm not sure what's wrong with this PR which supports the previously working pattern and actually fixes some of the inconsistent behaviors?

[tcleonard]

and actually, I think that #297 is not relevant anymore as it was already fixed (prior to this PR)