Creating a new issue based on lbryio/lbry-sdk#877

• Update ensure_directory_exists() to check that the directory is writable by the current process.
• If it's not, return a useful error
• Add tests to cover the case when
  • download directory does not exist
  • download dir exists but is not writable
  • dir exists an

Because in CliClient.resume() we don't account for the case, when there is an exception in one of the async processing stages:

void resume() {
	...
	try {
		client.startAsync(state -> {
			printer.ifPresent(p -> p.print(state));
			if (!options.shouldSeedAfterDownloaded() && state.getPiecesRemaining() == 0) {
				runtime.shutdown();
			}
		}, 1000); // <-- CompletableFuture.w

Currently we don't have any mechanism to limit the maximum number of clients that could be handled simultaneously.

This feature should be designed properly. Here is some clue: https://redis.io/topics/clients#maximum-number-of-clients

proposed by @mryab from #126

Virtual batching and LR scheduling are popular techniques with many applications. It would be nice to have an example of how to implement them with hivemind.

As of 0.8.15, ExpertBackend supports two ways of doing so:

• hacky way: by wrapping over optimizer and implementing LR schedule in opt.step
• orthodox way: by subclassing ExpertBackend and implementing ap

众所周知，tracker也是会被封的，所以每天获取有效的tracker能提高效率，至于tracker列表的获取，看这个
https://hao.su/3278/

we need to update the readme(s) to reflect the new deps