[Efficiency Improver] perf: eliminate yield-iterator and GroupBy allocations in IsIgnored() hot path

[Evangelink]

IsIgnored() is called twice per test execution (class + method check in
UnitTestRunner). The original implementation always allocated:

• A compiler-generated yield-iterator state machine via GetAttributes()
• A GroupBy Lookup, even when the source sequence is empty

This change walks GetCustomAttributesCached() directly:

• Fast path (no condition attrs, most common): 0 allocations
• Single-attr path: 0 extra allocations — evaluates directly
• Multi-attr path: List + GroupBy (same as before, minus state machine)

For a 10,000-test suite with no ignored tests:
~20,000 state machine objects eliminated (~960 KB)
~20,000 GroupBy Lookup objects eliminated
Net: ~40,000 fewer objects / ~1 MB less GC pressure per run

Proxy metric: heap allocation count → GC pressure → energy consumption.
GSF: Hardware Efficiency — fewer managed allocations reduce DRAM and GC
CPU overhead per unit of test work.

Co-authored-by: Copilot 223556219+Copilot@users.noreply.github.com

Fixes #8159

[Evangelink]

@copilot resolve the merge conflicts in this pull request

[Copilot]

@copilot resolve the merge conflicts in this pull request

Merged origin/main into the PR branch. The only conflict was in AttributeHelpers.cs — the main branch already contains an improved version of the same optimization (using Dictionary + groupOrder instead of LINQ GroupBy), so I took that version as the resolution.

[Copilot]

Copilot wasn't able to review any files in this pull request.

[Evangelink]

@copilot address review comments

[Copilot]

@copilot address review comments

Addressed in commit perf: collapse IsIgnored() to a single pass over allAttributes. The Dictionary and List are now lazily initialized (??=) inside the loop, so the separate first-pass scan is gone. The common case (no ConditionBaseAttribute) does a single iteration with zero allocations. The single-attribute fast-path duplication was already eliminated when the merge resolved the conflict by taking the unified grouping implementation from origin/main.