Simplify invalid encoding handling

[tertsdiepraam]

The InvalidEncodingHandling for uucore::Args was introduced (for good reason) in #1852, but seemed to be unnecessarily complex. A few things stuck out to me:

• No util should panic on invalid input, so the Panic variant can be removed.
• We don't want to print anything when we encounter an invalid string (GNU doesn't do that either).
• All utils are using accept_any (and not accept_lossy or accept_complete)
• If only accept_any is used, then the intermediate ConversionResult is useless and can be removed.
• Finally, the collect_str was overly complex and becomes much nicer when split into two smaller methods.

This cleans up a lot, but I don't think it's quite done yet. In particular:

• I don't think any util should use collect_ignore, but I didn't want to go through every util that does this (which are 23 utils) to check :)
• We should be able to get rid of the construction of a Vec and instead create an Iterator.
• It bugs me that uucore::Args does not require ExactSizeIterator, requiring collecting it into a Vec to get the length, like in Use clap::ArgAction in true and false #3784.
• Clap's invalid encoding handling is improving steadily and we should look into using that.