Guard against new positions with line or char less than 1

[zachallaun]

There was a report in the Elixir Language Discord of an error that occurred during a completion:

Error - 20:03:25] Request textDocument/completion failed.
  Message: ** (FunctionClauseError) no function clause matching in String.slice/3
    (elixir 1.17.2) lib/string.ex:2310: String.slice("    {:ok, assign(socket, form: form, email_empty = t), temporary_assigns: [form: form]}", 0, -1)
    (lx_common 0.5.0) lib/lexical/code_unit.ex:23: LXical.CodeUnit.utf8_position_to_utf16_offset/2
    (lx_protocol 0.5.0) lib/lexical/protocol/conversions.ex:146: LXical.Protocol.Conversions.extract_lsp_character/1
    (lx_protocol 0.5.0) lib/lexical/protocol/conversions.ex:120: LXical.Protocol.Conversions.to_lsp/1
    (lx_protocol 0.5.0) lib/lexical/protocol/conversions.ex:98: LXical.Protocol.Conversions.to_lsp/1
    (lx_protocol 0.5.0) lib/lexical/protocol/convertibles/lexical.document.edit.ex:11: LXical.Convertible.Lexical.Document.Edit.to_lsp/1
    (lx_lexical_shared 0.5.0) lib/lexical/convertible.ex:9: anonymous fn/3 in LXical.Convertible.Helpers.apply/2
    (elixir 1.17.2) lib/enum.ex:4858: Enumerable.List.reduce/3

  Code: -32603

This can only occur when a Lexical.Document.Position is created with :character set to 0. There is therefore an implicit invariant that any positions intended to be sent in a response have a positive character.

This PR makes that invariant explicit by adding a guard to Position.new/3 that will raise of any line/character that is not greater-than-or-equal-to 1.

There were a few places where we were using "utility positions" with line and/or character set to 0, but as far as I can tell, replacing these cases with line/character 1 does not change any behavior.

It is likely that there are still position-related bugs in Lexical, but this should help us catch them closer to where the bug is instead of during conversion to LSP where the stacktrace is pretty useless.

[zachallaun]

Agreed; added to 0.8 milestone.