[Performance] O(rules × references) nested loop in GetViolationsFrom for ProjectPath/ProjectTag detectors

## Description

`ProjectPathViolationDetector.GetViolationsFrom()` and `ProjectTagViolationDetector.GetViolationsFrom()` both iterate over all rules in the outer loop and all references in the inner loop, resulting in O(rules × references) complexity.

Unlike `AssemblyNameViolationDetector` (which has an experimental O(n) path using dictionary lookups), these detectors have no optimized path — the `GetViolationsFromExperimental()` just delegates back to the same nested loop.

## Affected Files

- `src/ReferenceCop/Detectors/ProjectPathViolationDetector.cs` — `GetViolationsFrom()`
- `src/ReferenceCop/Detectors/ProjectTagViolationDetector.cs` — `GetViolationsFrom()`

## Impact

For projects with many references (e.g., 50+ project references) and many rules, the nested iteration becomes costly, especially during MSBuild where this runs per-project.

## Suggested Optimization

**For ProjectTagViolationDetector:** Pre-filter rules that match the current project's tag (only one `fromProjectTag` can match). Then use a HashSet or dictionary keyed by `ToProjectTag` for O(1) lookup per reference:

```csharp
var matchingRules = this.rules.Where(r => r.FromProjectTag == fromProjectTag).ToList();
var blockedTags = new HashSet(matchingRules.Select(r => r.ToProjectTag));

foreach (var referenceContext in references)
{
var toTag = this.projectTagProvider.GetProjectTag(referenceContext.Reference);
if (blockedTags.Contains(toTag) && !referenceContext.IsWarningSuppressed)
{
var rule = matchingRules.First(r => r.ToProjectTag == toTag);
yield return new Violation(rule, referenceContext.Reference);
}
}
```

**For ProjectPathViolationDetector:** Similar approach — pre-filter rules matching the current project's path prefix, then check references linearly against only the matching subset.

Read GetViolationsFrom() and GetViolationsFromExperimental() in src/ReferenceCop/Detectors/ProjectPathViolationDetector.cs and ProjectTagViolationDetector.cs, then compare the experimental approach in AssemblyNameViolationDetector. Preserve rule matching and warning suppression while avoiding the nested rules-by-references iteration; done means both detectors retain the same violations with improved lookup complexity.

Written by the indexing model from the issue text.