Fix Enumerable.Sequence for NaN arguments and a zero step

[SimonCropp]

Bug

The Enumerable.Sequence polyfill (net7 to net9) diverged from the net10 BCL:

• A NaN start, endInclusive or step was not rejected. Every comparison with NaN is false, so the result was an endless sequence. The BCL throws ArgumentOutOfRangeException at the call, naming the offending parameter.
• A zero step was treated as a negative step, so Sequence(1, 5, 0) threw for endInclusive and Sequence(5, 1, 0) returned [5]. The BCL throws ArgumentOutOfRangeException for step unless start == endInclusive, in which case it yields the single value.
• The existing range checks passed the message text as the paramName ("endInclusive < start"). The BCL uses endInclusive.

Fix

Mirror the BCL: reject NaN up front, branch on step > 0 / step < 0, and handle step == 0 (including -0.0) separately. Exceptions now carry the same ParamName as the BCL.

Test

New Sequence_* tests for a NaN start, end and step, a zero and negative-zero step, a zero step with start == endInclusive, and the ParamName of the existing range checks. None of them enumerate, since the checks are not deferred. Five of the six failed on net8.0 before the fix. They pass on net8.0 and net9.0 (polyfill) and net10.0 (BCL). The expected behaviour was confirmed against net10 first. The net7.0 runtime is not installed on this machine, so net7.0 was compiled but not run.