ARROW-10610: [C++] Support exponential float nontion on big-endian in fast_float

[kiszk]

This PR is a follow-up of #8494 . The fast_float vendor library does not support big-endian of exponential float notion in from_chars. This PR correctly copies the available 32-bit (offset 4-7 on big-endian) to the result.

[github-actions]

https://issues.apache.org/jira/browse/ARROW-10610

[pitrou]

Did you report the issue upstream?

[kiszk]

Not yet. I will report this issue today.

[kiszk]

Reported at https://github.com/lemire/fast_float/issues/37

[lemire]

@kiszk Thanks for the report.

@pitrou How do you test Big Endian support? I have different ARM boards, but none of them run in Big Endian mode. I have not been able to find any publicly accessible CI system that run tests on Big Endian systems. I have a POWER system, but it is only little endian.

[lemire]

We are looking into fixing this upstream (big endian support).

[lemire]

(The hard part is to add the necessary CI testing.)

[lemire]

(I managed to add CI tests now.)

[lemire]

The issue has been fixed upstream (big endian support). I am running exhaustive tests and once they complete, I will issue a new release.

If you wish, you can wait for this new release instead of merging a vendored change.

[wesm]

Travis CI has s390x architecture (that's how we've been testing on BE)

[lemire]

@wesm Thanks to @kiszk, I figured that out.

That we could run CI on IBM mainframe-like systems was a shock to me. :-) I knew about POWER, but not that. So I learned something new today.

As written above, I have a patch that work. I'll issue a new release soon.

[lemire]

Version 0.2.0 of fast_float is out:

https://github.com/lemire/fast_float/releases/tag/v0.2.0

Release notes:

• Support for Big Endian systems
• Build with pedantic flags under GNU GCC without warnings
• More standard compliant handling of problematic strings (issue19)

[kiszk]

@pitrou Will you open another PR to integrate ver. 0.2.0 of fast_float?

[pitrou]

@kiszk I will, unless you beat me to it.

[kiszk]

@pitrou Sounds good, I will close this PR.