run_all.py: add --fast, --separate, and --close

[jprochazk]

Fixes #1945

• Refactored the run_all.py script
• Added --fast flag which limits examples to just cherry-picked examples
• Added --separate flag which runs each example in a separate viewer
  • For --web, this opens a new tab for each example
  • For native/--save, this opens a new window for each example
• Added --close flag which closes all open viewers when the examples finish running

The refactoring mainly consisted of moving the viewer starting logic into a class which can be used as a context manager in a with statement. It also finds and uses free ports every time it's started, which makes running it much more consistent (no more port already in use warnings).

Checklist

• I have read and agree to Contributor Guide and the Code of Conduct
• I've included a screenshot or gif (if applicable)

PR Build Summary: https://build.rerun.io/pr/2054

[jprochazk]

A few usage examples:

# Open fast examples, each in their own web viewer tab
$ ./scripts/run_all.py --fast --separate --web

# Open all examples sequentially in the same viewer
$ ./scripts/run_all.py

# Skip building the SDK, open fast examples, save them to .rrd files, then load each one in a separate window
$ ./scripts/run_all.py --skip-build --fast --save --load --separate

# Skip building the SDK, open fast examples sequentially, loaded from `out.rrd` files
$ ./scripts/run_all.py --skip-build --fast --load

[emilk]

When I run ./scripts/run_all.py --fast --separate I get at least one window with

Running ./scripts/run_all.py --fast --separate --web gets me several tabs with the same message.

I haven't reviewed the code yet, but seems to me there is a place or two where we need to add a wait.

For instance, we need need to get the web browser time to connect to the rerun server (and download the log data) before shutting it down. The easy way here is to give it 5-10 seconds or something. The robust solution is quite a bit more involved, as it involves figuring out if the rerun process has both received the full log stream and also served it on in re_ws_comms::RerunServer. I suggest we punt on that.

[emilk]

Really nice, but there are some weirdness, at least on my Mac!

I would also love for the text output of run_all.py to be improved. Especially when running in sequential mode it would be good if each example was clearly delineated and marked, e.g. with a blank line and a `print("\nRunning {args}…")

[emilk]

Any non-zero returncode should result in run_all.py returning non-zero, and printing a big fat error message

[jprochazk]

Changed this to assert instead. Not sure how I should do the same when running --separate.

[emilk]

Works great! ⭐