Fix aggregation parsing for complex expressions

[nicosuave]

Summary

• Replace greedy regex with sqlglot in Metric class to properly handle expressions like SUM(x) / SUM(y) without mangling them
• Fix COUNT DISTINCT detection using isinstance(parsed.this, exp.Distinct)
• Add expression metric support in SQL generator for metrics with inline aggregations
• Fix dependency analyzer to skip resolution for expression metrics with inline aggregations
• Fix cumulative metrics to properly resolve references to other measures and generate valid aliases
• Add kitchen sink tests using patterns from rill-examples to catch edge cases

Bugs Fixed

1. Greedy regex: SUM(deletions) / SUM(changes) was being parsed as agg=sum, sql=deletions) / SUM(changes
2. COUNT DISTINCT: COUNT(DISTINCT user_id) was parsed as agg=count, sql=DISTINCT user_id instead of agg=count_distinct, sql=user_id
3. Expression metrics: Metrics with full SQL expressions (no agg, no type) couldn't be queried
4. Window functions over measures: Cumulative metrics referencing other measures failed to resolve

Test Results

• 42/42 Rill adapter tests pass
• 25/25 new kitchen sink tests pass

[chatgpt-codex-connector]

💡 Codex Review

Here are some automated review suggestions for this pull request.

Reviewed commit: 11a079a1d9

ℹ️ About Codex in GitHub

Codex has been enabled to automatically review pull requests in this repo. Reviews are triggered when you

• Open a pull request for review
• Mark a draft as ready
• Comment "@codex review".

If Codex has suggestions, it will comment; otherwise it will react with 👍.

When you sign up for Codex through ChatGPT, Codex can also answer questions or update the PR, like "@codex address that feedback".

[chatgpt-codex-connector]

Preserve all DISTINCT expressions in COUNT

When COUNT(DISTINCT ...) has multiple expressions (e.g., COUNT(DISTINCT a, b)), sqlglot exposes them in Distinct.expressions. This code now only takes expressions[0], which drops the remaining expressions and changes the aggregation semantics (distinct on a instead of (a,b)), so counts will be too high for multi-column DISTINCTs. The previous regex path preserved the full a, b list, so this is a regression for any multi-column distinct usage.

Useful? React with 👍 / 👎.