Detect the error case where `scp` cannot copy to remote because remote doesn't have `scp` installed

[joshspicer]

As discussed here, to use scp it needs to be installed on both sides. Today the ssh extension will not be able to identity this fatal case, likely producing a cryptic error like:

[17:11:53.801] > bash: line 1: scp: command not found
> lost connection

More context: microsoft/vscode#226759 (comment)

[chrmarti]

Not sure if this could be applied here, but Dev Containers uses an existing shell to send the server over. It only requires dd and tar on the remote side.
https://github.com/microsoft/vscode-remote-containers/blob/7457cd00ff79e5bcf1ef50c8028fd44edab60a2e/src/common/injectHeadless.ts#L1065

[joshspicer]

Interesting, i'm not able to repro with a test Ubuntu 24 remote that has scp removed. This behavior must only exist in some cases of SSH

Edit: by "removed" I break the binary on the path (by changing /usr/scp to /usr/scp-BROKEN)

[joshspicer]

I can't repro this anymore (both with the aforementioned dev container repro or a real VM). Perhaps I was actually hitting a bug/inconsistency in fedora at the time? Not sure - please do comment if you can reproduce today.