[C#] Improve "isExponentialRegex" detection logic in ReDoSQuery.qll to prevent false negatives

Hello,

In the C# security analysis suite, the query `Denial of Service from comparison of user input against expensive regex` (`cs/redos`) relies heavily on underlying helper logic to flag regular expressions with potential exponential behavior. Specifically, in `csharp/ql/lib/semmle/code/csharp/security/dataflow/ReDoSQuery.qll` (lines 58–72). This uses a set of hardcoded regular expressions via `regexpMatch` to identify string literals that represent exponential (ReDoS-vulnerable) regular expressions.

While these three variations catch patterns like `([a-z]+.)+`, they are fragile syntactic approximations. This approach misses variations of overlapping or nested quantifiers, creating a scenario where dangerous regex structures easily bypass the query's detection due to minor structural variations.

For example the query overlooks risky patterns like these:
- Nested Quantifiers without literals: `(a*)*b or (x+)*`
- Overlapping Alternations/Sequences: `(x+x+)+y`
- Complex or Distant Structural Paths: Patterns that contain non-trivial prefixes/suffixes or specific character class structures can fail to match the strict capture-group structures defined in the QL code. Depending on the engine evaluating these meta-regexes, they could themselves face performance degradation when scanning highly complex, non-matching input paths.

This issue stood out because these specific pattern variants can easily slip through the ReDoS query undetected. This creates a gap between the security results and the actual risk. I'm wondering if it would be possible address this in a future version?

Version: 2.26.0

Comece com csharp/ql/lib/semmle/code/csharp/security/dataflow/ReDoSQuery.qll, especialmente as linhas 58–72, e rastreie como a consulta cs/redos usa isExponentialRegex e regexpMatch. Compare a lógica de matching existente com os exemplos listados de quantificadores aninhados e alternativas sobrepostas. Considera-se concluído quando a consulta detectar essas variantes sem perder a cobertura existente nem introduzir um comportamento de matching problemático.

Escrita pelo modelo de indexação a partir do texto da issue.