Complete more precisely in the "value" position

[rolandwalker]

Description

When in the "value" position, that is, where a value may be referred to such as a literal, function name, or column name, don't offer all keywords, but instead limit keywords to function names and function-alikes.

Pygments has some errors in its designations of keywords vs functions, such as classifying JSON_VALUE() as a keyword, and some missing functions as well, so we amend the values imported from Pygments. At a certain point the amendments would be large enough that we should consider maintaining our own categorical lists. But the amendments are so far not too extensive.

(We might also consider adding loadable function names.)

Since we have made the list of function names more accurate, we can then remove {"type": "keyword"} from the completion candidates when we are in the "value" position (keeping {"type": "function"} which is already present.

Now, functions such as JSON_VALUE() complete in the "value" position, eg, after a SELECT, but mere keywords such as SELECT do not. We no longer suggest SELECT SELECT!

An exception was made here for completions within backticks, which are still not great, and need future work, because the choices are too many. Backtick completions are left in the current state.

As a comment notes, we should better also define "value position" keywords such as CASE and make a separate completion set for them, rather than lumping them into the list of functions as is done here.

There are other edge cases such as CURRENT_TIME, which can occur in the value position but does not take parentheses, and weird ones such as MEMBER OF(), a midfix function which also contains a space in the name.

Checklist

• I added this contribution to the changelog.md file.
• I added my name to the AUTHORS file (or it's already there).
• To lint and format the code, I ran

  uv run ruff check && uv run ruff format && uv run mypy --install-types .

[github-actions]

1. Correctness regression: value-position literals/operators are no longer completable

• In the select/where/having branch, non-backtick value positions now return only column, function, and alias, with keyword removed (completion_engine.py:362, completion_engine.py:385).
• This drops completion for valid value-position tokens that are not functions, e.g. NULL, TRUE, FALSE, NOT, BETWEEN, etc. (only CASE was re-added indirectly via functions) (sqlcompleter.py:836).
• Action: keep a dedicated “value-position keywords/literals” set (separate from function names) and include it in this branch.

2. Test gap: no guard for the above regression

• Updated tests removed keyword expectations broadly but do not add targeted assertions for value-position literals/operators (e.g. SELECT NU, WHERE col = NU, WHERE NO) (test_smart_completion_public_schema_only.py:210, test_completion_engine.py:21).
• Action: add focused tests for value-position non-function tokens to prevent future regressions.

No security issues found in this PR.

I could not execute tests in this environment because pytest/uv are unavailable.