Add workflow to catch broken URLs in the docs

[jatkinson1000]

Spotted by @niccolozanotti and patched here.

I also did a quick scan for anything else that may be outdated.

[jatkinson1000]

Tried:

• https://github.com/gaurav-nelson/github-action-markdown-link-check
  Struggles to parse internal links, and reports 403 errors for a number of intel webpages see 8a67b93
• https://github.com/RazcoDev/broken-urls-in-code
  Struggles with extracting links properly due to markdown syntax (trailing ) etc.) That was not easily fixed for all edge cases with modification to the regex - see 1760a89

Will leave for now as non-urgent and taken a fair amount of time today.

[jatkinson1000]

Made my own 🙄

Works fine on my machine, but fails on GitHub CI for the gcc links.
Returns 000 exit code for curl.

Is Gnu blocking requests from GitHub servers?? 🤣

This was supposed to be simple 😭

[jatkinson1000]

Huh, updated but passed fine that time 🤷

[jatkinson1000]

TODO:

• Add some docs
• Rename to checkmdlinks.sh

[jatkinson1000]

@Mikolaj-A-Kowalski any thoughts appreciated here.

This seems rather fickle (rate limiting and bots/redirects to be checked manually) to run as part of CI.
I'm wondering if we just add it as a script and rely on developers to run instead as a useful utility - see #574.

Further, there are some links to TorchScript documentation that it seems PyTorch has simply deleted from the latest version.

[Mikolaj-A-Kowalski]

I would not be inclined to scrap the automated check completely (as in do we really trust ourselves to run the script as often as we should ;-) ), but I completely agree that it is far to brittle to be blocking for the CI.

What I am thinking is we could add continiue-on-error to the link check step so it doesn't kill the job.

Also to make the results more visible we can raise a 'notice' using the appropriate workflow command, i.e. for the failure echo something like that:

echo "::notice title=Maybe broken link:: '$url' may not be reachable got  (HTTP $status_code). Note this check is brittle. Please check manually and ignore this message if the link seems OK."

This should create a helpful (?) banner. Not entirely sure where though if the file and line is not explicitly specified...

May also be a better idea to collect all failing links into a single message to avoid getting multiple banners.

My thinking is that we can keep the check running for some time and we will get a sense of how annoying vs useful it will be. If it creates too much noise we can always scrap it.

[joewallwork]

This version of the script mostly works well for me, although I get the rate limit errors mentioned.

I'm wondering if we just add it as a script and rely on developers to run instead as a useful utility

How about we add it as a CI job whose only trigger is calendar-based? e.g., runs early AM on the first day of the month?

[joewallwork]

Is it bad practice to ship scripts that are already executable?