question: why schedule observer per-directory?

[crockeo]

Hi! This isn't an issue request so much as a discussion, but I had a question about the implementation that I wanted to ask about.

In jurigged.live::Watcher::on_prepare we end up self.observer.schedule-ing once per-directory:

jurigged/src/jurigged/live.py

Lines 96 to 98 in afed805

	def on_prepare(self, module_name, filename):
	JuriggedHandler(self, filename).schedule(self.observer)
	self.registry.log(WatchOperation(filename))

->

jurigged/src/jurigged/live.py

Lines 150 to 154 in afed805

	def schedule(self, observer):
	# Watch the directory, because when watching a file, the watcher stops when
	# it is deleted and will not pick back up if the file is recreated. This happens
	# when some editors save.
	observer.schedule(self, os.path.dirname(self.filename))

This can run into OS-level limits on the number of file watchers available, which causes jurigged to fail on large projects. I've been messing around with this on such a project, and found a workaround by monkeypatching Watcher instead use a single watchdog.Observer.

I was wondering: is this a bad idea for a reason I don't understand yet? And if not: any chance y'all would accept a pull request to make this change?