Add `backports.datetime_fromisoformat` to the benchmarks

[movermeyer]

What are you trying to accomplish?

Fixes #136.

What approach did you choose and why?

At first, I simply add it as another library like any other.

backports.datetime_fromisoformat is fast, thanks to it using the same lightweight time zone implementation as ciso8601, which means that it is actually faster than datetime.

This means that it will likely always be 2nd place on the chart. But it also means that as written it will always be the library of comparison:

Which seems a little awkward, since backports.datetime_fromisoformat will (by definition) never apply to the latest version of Python.

Ideas I considered:

• We could exempt backports.datetime_fromisoformat from ever being selected for comparison. In this case, it would compare datetime instead.
  • But this is confusing, since backports.datetime_fromisoformat is clearly the second row.

• backports.datetime_fromisoformat and datetime have a special relationship, and don't overlap in the versions of Python that they support. They could be merged into a single line in the table

  • I tried it. I was alright:
  
  • If a hypothetical future Python x.y extends the supported subset of ISO 8601 once again, and backports.datetime_fromisoformat backports that version, then they will overlap in the supported versions of Python. 🤷
  • Since backports.datetime_fromisoformat is faster, you can see a sudden spike in runtime from 3.10 -> 3.11. That's kinda confusing I guess. 🤷
  • Ultimately, it was getting too close to misrepresenting datetime for my liking
• We could simply not include backports.datetime_fromisoformat at all
  • It could be similar to hardcoded, where we do the benchmarking for it, but then don't include it in the table.

After lots of fussing with it, I realized that I could simply change the comparison summary line.

I changed it so that it will only compare against libraries that support the latest version of Python, and updated the text to match:

ciso8601 takes 97.2 nsec, which is 1.8x faster than datetime (builtin), the next fastest Python 3.11 parser in this comparison.

What should reviewers focus on?

• What do you think about the end result?
  • You can see the rendered version here

The impact of these changes

backports.datetime_fromisoformat is benchmarked and represented in the table.

[movermeyer]

I don't see the README updated in this change, will that come in a following PR?

No CHANGELOG entry is needed. Benchmarking scripts are not for end consumers, just for the generation of the performance charts in the README.

[AlecRosenbaum]

I didn't mean the changelog, I meant the readme (the screenshot from the PR description / linked readme.rst file?)

[movermeyer]

Oh yeah, that'll change will happen when I refresh https://github.com/closeio/ciso8601/pull/135