Handle surrogate pair when converting from LSP encoding

[bullno1]

The generic LSP completer adds 1 to the code unit position before conversion to get a 1-based index:

ycmd/ycmd/completers/language_server/language_server_completer.py

Lines 3489 to 3490 in a51329a

	lsp.UTF16CodeUnitsToCodepoints( line_value,
	location[ 'character' ] + 1 ) )

However, this is problematic when the original index is pointing at a surrogate pair.
For example: given the string: 😊, the language server sends index 0.
Adding 1 will point it to the low surrogate and the following conversion will fail with a UnicodeError:

ycmd/ycmd/completers/language_server/language_server_protocol.py

Lines 826 to 827 in a51329a

	bytes_included = value_as_utf16_bytes[ : code_unit_offset * 2 ]
	return len( bytes_included.decode( 'utf-16-le' ) )

This is because the surrogate pair is sliced in half.

What this PR does is checking whether the high byte at the offset is a low surrogate and advance the offset further to skip the entire pair.

This only happens for positions sent from the language server so it is not a reversible conversion.
I added a separate test case from test_CodepointsToUTF16CodeUnitsAndReverse

---

This change is 

[puremourning]

thanks!

Reviewed 2 of 2 files at r1, all commit messages.
Reviewable status: 1 of 2 LGTMs obtained (waiting on @bullno1)

[puremourning]

@Mergifyio rebase

[mergify]

rebase

✅ Branch has been successfully rebased