maint: Replace use of folly getCurrentThreadId with STL

[ianthomas23]

Reference Issues/PRs

What does this implement or fix?

This is work to assist @jjerphan in removing use of folly in the ArcticDB code base.

The change here is to remove use of folly::getCurrentThreadID() in get_thread_id(), replacing it with use of std::this_thread::get_id().

This replaces #1408 and is submitted from a branch of the main repo rather than my fork so that CI runs automatically.

Any other comments?

The folly implementation is much more complicated as there are different implementation for the 3 major OSes:

https://github.com/facebook/folly/blob/69a1b93df285457cd0799a6898e2e5390a09cd73/folly/system/ThreadId.cpp#L29-L37

The different implementations are not required as the STL C++11 solution should work on all OSes. Following
#1408 (comment) the solution here returns a std::thread::id rather than a uint64_t and this is formatted correctly in log messages (the only use of thread ID) by including <fmt/std.h>.

Checklist

Checklist for code changes...

• Have you updated the relevant docstrings, documentation and copyright notice?
• Is this contribution tested against all ArcticDB's features?
• Do all exceptions introduced raise appropriate error messages?
• Are API changes highlighted in the PR description?
• Is the PR labelled as enhancement or bug so it appears in autogenerated release notes?

[ianthomas23]

Some CI builds are passing and some failing. The failures are

error: invalid use of incomplete type ‘struct std::variant_size<long unsigned int>’
     struct variant_size<volatile _Variant> : variant_size<_Variant> {};
            ^~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~

I can reproduce this locally using a docker build whereas a conda build works fine (on Ubuntu 13.10).

I've tracked this down to fmtlib/fmt#3068 which was fixed in fmtlib/fmt#3072, dated 2022-09-03. The first release of fmt to include this fix is 10.0.0. I note that

ArcticDB/environment_unix.yml

Lines 42 to 43 in baf9eb9

	# ArcticDB is currently incompatible with fmt 10
	- fmt < 10

indicates we need to stick with fmt < 10.

Apparently gcc >= 11 doesn't need the fmt fix which is presumably why building using conda works OK (I have gcc 12.3.0 this way) whereas in the docker image I have gcc 10.2.1.

I suppose the easiest way to avoid these problems is to remove the #include <fmt/std.h> and instead locally implement a formatter for std::thread::id.

[jjerphan]

We might be able to use fmt 10.

[ianthomas23]

We might be able to use fmt 10.

I will take a look.

[jjerphan]

LGTM given 🟢 CI checks.

Thank you for helping with #1412, @ianthomas23.