[tests] Keep GetObjectArray peer alive

[jonathanpeppers]

JnienvArrayMarshaling.GetObjectArray intermittently compared two managed wrappers for the same Java context because the wrapper inserted into the JNI array was not explicitly retained. Java peer registration uses weak references, so this could fail under CoreCLRTrimmable GC timing even though JNI marshaling returned the correct Java object.

Capture Application.Context in a local, use that instance to construct the JNI object array, and assert that the roundtrip returns the same known-live managed peer. This preserves the intended peer-identity coverage without requiring a runtime change.

• Useful description of why the change is necessary.
• Links to issues fixed
• Unit tests: source diff and whitespace validation passed. Focused on-device validation was unavailable because the worktree has no local .NET for Android SDK build or adb.

[Copilot]

Pull request overview

This PR fixes an intermittent test failure in JnienvArrayMarshaling.GetObjectArray caused by managed peer identity instability: the test could end up comparing two different managed wrappers for the same Java Context because the wrapper used to populate the JNI object array wasn’t explicitly kept alive (peer registration uses weak references).

Changes:

• Capture Application.Context into a local variable and reuse it when constructing the JNI object array.
• Assert round-tripped peer identity against the known-live local context instance instead of a fresh Application.Context call.

[jonathanpeppers]

@dalexsoto review

[dalexsoto]

No blocking issue remains: the test-only lifetime change is harmless on the current runtime path, and the underlying peer-replacement fix is already present.