[Wip] Return sensible error in integer-decode-float when called with a infinity value

[kpoeck]

• issue encountered while cross-compiling sbcl (and the origin is still to be clarified)
• return sensible error instead of sigfpe as before
• added test to regression-suite

[Bike]

Preferably we should figure out how to transmute a SIGFPE into a floating point error, because otherwise we'd have to put checks like this on every floating point operation. I thought I had it doing that already, but evidently not. Either the error isn't one of the ones we handle, or the traps are just totally messed up. Code is here: https://github.com/clasp-developers/clasp/blob/dev/src/gctools/interrupt.cc#L256-L271

[kpoeck]

I hope in #955 I have this fixed as you intended

[Bike]

yes. is this PR dead, then? incorporating the tests would be good, but #961 precludes that.

[kpoeck]

As you said, this pr is dead. I have stored the tests locally, but as you said, we can't add them right now. I wonder whether we need something like (ext:issue-fixed-p :961)