[tobias-tengler]

Capturing pocketed errors creates a list wrapper even though completion callers only enumerate the eager snapshot.

This PR returns a fresh eager array directly from the snapshot helper.

Entry order, shared path/error references and snapshot isolation during callbacks are preserved; focused tests pass on .NET 10 and .NET 11.

Release .NET 10 measurements of snapshot capture plus a common indexed reference consumer against pinned main 15888585e1 show stable three-error allocation reduction from 104 to 72 B (32 B, 30.77%). The accepted elapsed costs are 12.71% (3.88 ns) for three errors and 6.61% (1.11 ns) for one error; no elapsed improvement is claimed. One-error allocation readings were 72/72→0/40 B across repetitions, so the displayed 72→20 B arithmetic mean is uncertain and supports no allocation claim. The empty guard is a control.

Method	Version	Mean	Ratio	Allocated	Alloc Ratio
Empty guard control	main	0.861 ns	1.00	0 B	N/A
Empty guard control	PR	0.596 ns	0.69	0 B	N/A
One pocketed error snapshot	main	16.783 ns	1.00	72 B	1.00
One pocketed error snapshot	PR	17.893 ns	1.07	20 B	0.28
Three pocketed errors snapshot	main	30.496 ns	1.00	104 B	1.00
Three pocketed errors snapshot	PR	34.372 ns	1.13	72 B	0.69

[github-actions]

Patch coverage

100.0% of changed lines covered (1/1)

File	Covered	Changed	Patch %
…/src/Fusion.Execution/Execution/Results/FetchResultStore.cs	1	1	100.0% 🟢

---

Project coverage: 59.0% (311445/527757 lines)