FIxes #5380

By adding classes that clearly define when a hint is a table or a keyword, it gives developers easy control if they wish to style hints differently. For example, you may wish to style table hints in a different colour to keywords (to make them easier to distinguish).

Issue #5179

The merge view will incorrectly show a "Revert chunk" arrow when
the text on both sides of the view are the empty string.

The getChunks() helper is not prepared to deal with an empty diff
array.

The getDiff() helper returns an empty diff array when both strings
to compare are empty, as the only diff item returned by
diff_match_patch, `[0, ""]`, is removed as a result of the diff
entry containing an empty string.

The chosen (trivial) fix is to simply check for the empty diff
array case at the top of getChunks().

Issue #5312

The hint logic of the `fromList` command did not respect the cursor position inside the token.
Now it will list all matching completions for the substring up to the cursor, instead of only the ones matching the complete token. The behavior is now the same as for the regular autocompletion.

Closes #5290

When using the editor in "application/json" mode, the standard javascript autocompletion would show up once triggered. Although it's deactivated by default and doesn't make sense to enable autocompletion in JSON mode on its own, this change fixes incorrect completions once you combine the default completion with your own custom logic.

Primary motivation was to remove the use of `innerHTML` because the
Firefox extension validator raises a warning about this.

The obvious solution is to use `textContent` instead of `innerHTML`
but I thought it could be useful to have the ability to pick a
custom visual for the scroll-lock icon.

As per MDN documentation, all the supported browsers as per
CodeMirror documentation support `:after`.

Issue #5265

This explicitly adds keywords that are already syntax-highlighted
and correctly parsed by CodeMirror already in the JavaScript mode.

[javascript-lint] JSHINT backward compatibility - Setting error severity to error by default when there is no error.code property (#5260)

Closes #5238

When using tabs for indentation, the markers would not always align with the actual problem.
The old workaround for JSHint never really worked (only when a line had one tab) and I found a simple option which fixes the issue.
This time I also checked backwards compatibility, which works even fine with the old JSHint version R07.
As a second marker fix, I added an error line check, since JSHint for some reason adds configuration issues in the errors array with line: 0, character: 0. Now, if you pass an invalid option you will get a console error instead of an invisible marker.

When enabling strict equality checks via `lint: {options: {eqeqeq: true}}`, found problems showed up as errors instead of warnings.
To fix it, the `fixWith()` logic had to be changed since simply adding the phrase to the warnings array would not have worked. This is because both the warnings and errors array matched this exact error and therefore the severity could never be "warning" (errors were checked after warnings).

I didn't include "Missing property name" and "Unmatched " in the new error array since they already are errors with the new logic.
"Stopping, unable to continue" also got removed since it didn't appear anywhere in the current jshint.js.

For now I've implemented everything to not break previous behavior/hinting, except the strict equality hint severity. Although I want to suggest removing the following codes from the new error array (so they can stay warnings):
- W033 (Missing semicolon) - since erroneous missing semicolons have their own code: E058
- W084 (Expected a conditional expression and instead saw an assignment) - since something like `switch (var2 = var1 + 42)` is valid js code, though not recommendable
- maybe W023/24/30/90, since there are many more " and instead saw an" hints that are already errors with their own codes, so I think they should be pretty accurate. Unfortunately I couldn't force these warnings so I couldn't check.

No longer creates a fold spot for both lines of a line-broken
tag.

Closes #5179

Issue #5177

If you try to autocomplete at line 0, column 0 after previoulsy having edited for example a WHERE clause, the autocompletion triggers with an invalid position, since prevItem is null.
Maybe my fix isn't the best solution, and you could even avoid to enter findTableByAlias enitrely, I don't know.

Show column hints (if a defaultTable is set) above table hints, since
you are far more often in clauses where you need those.

Closes #5174

Closes #5106

Closes #5095

Issue #4792

Stop relying on mode tokens, use simple heuristic of not being after a word char

Issue #5058
Issue #2657

Closes #5059

Per #5030, when adding new items to the middle of a Markdown list, the remaining list numbers should automatically increment

Issue #5054

To also make sure the last token isn't a string.

Issue codemirror/google-modes#74

The markdown mode uses `CodeMirror.innerMode` which isn't defined in
the stripped down node runmode. Since markdown is the only mode
(AFAICT) that uses it I'm not sure if it would make more sense rewrite
it to not use it but this seemed like the least risky option.

Issue #5031