[hailctl] batch submit fixes

[ehigham]

CHANGELOG: Fix many issues, including (hail#14274), with hailctl batch submit introduced in 0.2.127.
Fixes #14274, Replaces #14351 (authored by @jigold) and #14186 (authored by @danking).
This change has low security impact

Dan's original commit message is as follows:
"""
Issues resolved herein:

1. build.yaml tests must not use exit 0 as it exits the test early.
2. Always prefer orjson to json.
3. Add --wait which waits for the submitted batch to complete and exits success only when the batch is success.
4. Whenever working with paths, we must use the realpath which resolves symlinks. In particular, on Mac OS X, /tmp is a symlink to /private/tmp and Python's APIs are inconsistent on whether they return a realpath or a path with symlinks. [1]
5. If the destination looks like a directory (e.g. "bar:/foo/", "bar:/"), the tests all suggest we should copy into not to. We now check for a trailing slash and copy into.
6. ln -s src dst means different things depending on whether dst is an extant folder or not. In this PR, I prefer to always be fully explicit so I never rely on ln detecting the destination is a directory and acting differently. Put differently: file_input_to_src_dest now never returns a file source and a destination folder.
7. We need to create the real_absolute_cwd() on the job before we cd into it.
8. test_dir_outside_curdir suggests that --file foo/:/ is meant to copy the contents of foo into the root. This cannot be implemented with our symlink strategy (you can't replace the root with a symlink), so I changed the interpretation: a trailing slash on the source is meaningless. If the destination ends in a slash, we "copy into", otherwise we "copy to".
9. Add examples of --files usage.

[1]:

In [1]: import tempfile
   ...: tempfile.TemporaryDirectory()
Out[1]: <TemporaryDirectory '/var/folders/x1/601098gx0v11qjx2l_7qfw2c0000gq/T/tmp_pmj3lr9'>

In [2]: import os
   ...: os.getcwd()
Out[2]: '/private/tmp'

In [3]: !ls -al /tmp
lrwxr-xr-x@ 1 root  wheel  11 Aug  2 05:44 /tmp -> private/tmp

"""

[ehigham]

This change is @jigold's edits manually applied on main.
I couldn't figure out the merge conflicts when rebasing #14351.
Mixing merges and rebases causes havoc!

[chrisvittal]

Looks good. Thanks for pushing this over the line.

[cjllanwarne]

This seems well implemented and the test cases give me good confidence. My only minor concern is relying on a trailing '/' to decide whether something is a directory or not. If you can make me feel better about that, I'll happily 👍 . Oh, and I think a minor documentation improvement is possible too 😄

[cjllanwarne]

Should we be adding a / to the end here to make cwd look like a directory?

[ehigham]

As you can tell, this sort of thing is unreliable and inconsistently applied. The short answer is no. I'm gonna change this implementation to use pathlib for local files.

[cjllanwarne]

Is this really the best we can do? Can we not make an os call to find out for sure whether or not it's a directory?

[ehigham]

I'm going to take a detailed look at this PR now. I initially just assumed it was good to go but it's clear that's not possible.

[ehigham]

This was mostly implemented by @danking - I just tweaked the tests to not drool invalid current working directories.
I hate trialing slashes to denote directories. If we're doing that anywhere here I'll gladly kill it.