[Native] Remove unused field weak_handle

[AustinWise]

This field on JniObjectReferenceControlBlock was never written to. The only time it was read was for debug logging.
This shrinks the size of the control block from 32 to 16 bytes on 64-bit platforms.

While I was at it, I added the the StructLayout attribute to the JniObjectReferenceControlBlock in C#. This makes the warning suppression for CS0649 no longer necessary.

---

• Useful description of why the change is necessary.
• Links to issues fixed
• Unit tests

[jonathanpeppers]

/review

[github-actions]

✅ Android PR Reviewer completed successfully!

[github-actions]

🤖 Review — ✅ LGTM (minor)

Clean, well-scoped removal of the unused weak_handle field from JniObjectReferenceControlBlock. Verified independently before reading the description.

Verification performed:

• 🔎 weak_handle has zero remaining readers or writers anywhere in the tree — confirmed via full-repo search.
• 🧱 All four struct definitions are updated in lockstep, keeping the field layout identical across the managed/native boundary:
  • external/Java.Interop/.../JniObjectReferenceControlBlock.cs
  • src/Mono.Android/.../JavaMarshalRegisteredPeers.cs (internal mirror)
  • src/native/clr/include/host/gc-bridge.hh (CoreCLR)
  • src/native/mono/monodroid/osbridge.hh / osbridge.cc (Mono)
• 📐 Resulting layout is handle (8) + handle_type (4) + refs_added (4) = 16 bytes on 64-bit, matching the PR's claim. bridge-processing.cc only touches handle/handle_type/refs_added, so the CLR bridge is unaffected.
• 🪵 The two osbridge.cc diagnostic log sites that referenced key_handle (derived solely from weak_handle) are correctly cleaned up, including the format strings — no dangling format specifiers.
• ✅ Dropping the #pragma warning disable CS0649 is now valid since the mirror struct is explicitly assigned/read via native memory and the previously-unassigned field is gone.

Notes:

• 💡 One optional consistency suggestion posted inline re: adding the explicit [StructLayout] to the source Java.Interop struct as well.
• PR is a draft and mergeable_state is blocked; the description leaves "Unit tests" unchecked. This is a pure field/logging removal with no behavioral change, so a dedicated test isn't strictly warranted — but confirm the native GC-bridge builds (Mono + CoreCLR) pass in CI before merging, since this touches the managed↔native struct contract.

Severity counts — ❌ 0 · ⚠️ 0 · 💡 1

Generated by Android PR Reviewer for #12239 · 64.4 AIC · ⌖ 18.6 AIC · ⊞ 6.8K
_{Comment /review to run again}

[github-actions]

🤖 💡 Consistency — Good call adding [StructLayout (LayoutKind.Sequential)] here to make the native-layout dependency explicit and drop the CS0649 suppression. For symmetry, consider adding the same attribute to the source Java.Interop.JniObjectReferenceControlBlock struct, since the native C++ definitions (gc-bridge.hh, osbridge.hh) read both by field offset. C# defaults structs to sequential layout so this isn't a bug today, but making it explicit on both mirrors documents that the ordering is intentional and load-bearing.

(Rule: Interop struct layout must match native definitions)

[AustinWise]

I tried to keep these changes as minimal as possible, since .NET 11 is probably branching soon. Also, the definition of Java.Interop.JniObjectReferenceControlBlock was recently merged into this repository from the old Java.Interop repo. I'm assuming that duplicate copy is on the way out.

Anyways, I was not sure where to draw the line on these changes. Let me know if you want any revisions.