WIP s3 loading.

[pp-mo]

Closes #6374

This approach uses f3fs-FUSE

Tested: have shown that this can load netcdf and PP files from an S3 bucket.

Much to-do

• add f3fs-FUSE to deps
• lots of tests
• consider saving
• further explore the "unmount problems"

[codecov]

Codecov Report

❌ Patch coverage is 5.76923% with 98 lines in your changes missing coverage. Please review.
✅ Project coverage is 89.93%. Comparing base (8ed2570) to head (ada62dd).
⚠️ Report is 6 commits behind head on main.

Files with missing lines	Patch %	Lines
lib/iris/io/s3_mounts.py	0.00%	92 Missing ⚠️
lib/iris/loading.py	50.00%	5 Missing and 1 partial ⚠️

Additional details and impacted files

@@            Coverage Diff             @@
##             main    #6731      +/-   ##
==========================================
- Coverage   90.29%   89.93%   -0.36%     
==========================================
  Files          91       92       +1     
  Lines       24472    24577     +105     
  Branches     4569     4583      +14     
==========================================
+ Hits        22097    22104       +7     
- Misses       1607     1704      +97     
- Partials      768      769       +1     

☔ View full report in Codecov by Sentry.
📢 Have feedback on the report? Share it here.

🚀 New features to boost your workflow:

• ❄️ Test Analytics: Detect flaky tests, report on failures, and find test suite problems.

[pp-mo]

Update: @ukmo-ccbunney pointed out that it could be simpler to leave "s3fs" mounting to the user.
That certainly looks like a simpler approach, and doesn't mire us in some complex issues which this approach raises:

• testing is not too simple
• introduces awkward dependencies
• could be tricky to maintain
• apparently, works for Linux/macOS/FreeBSD -- but not elsewhere
• the indeterminate unmount problem ("umount target is busy") is a worry -- especially if it could have security implications

What we might do is provide some simple advice in the Iris docs under "Further Topics" ?