What changed

• add a focused JIT regression for ref GLOBAL_TEE overwrite behavior in interp/interp_test.go
• cover the changed globalPut(..., pop=false) path introduced by recent ARM64 JIT refcount work

Paths tested

• interp/jit_arm64.go: JIT GLOBAL_TEE ref overwrite path
• interp/interp.go: overwritten ref release remains observable through runtime RC state
• interp/interp_test.go: new regression in TestInterpreter_JIT

Added tests

• TestInterpreter_JIT/compiles_ref_global_tee_and_releases_overwritten_ref
  • compiles GLOBAL_TEE for a ref global
  • verifies the overwritten ref is released
  • verifies the new ref remains live and equal through GLOBAL_GET + REF_EQ

Validation

• go test ./interp -run 'TestInterpreter_JIT/compiles_ref_global_tee_and_releases_overwritten_ref|TestInterpreter_JIT/compiles_ref_global_get|TestInterpreter_Run' -count=1
• go test ./interp -count=1

Skipped targets

• GLOBAL_SET ref JIT overwrite: skipped because numeric globals already cover the pop=true store shape; this patch targeted the untested GLOBAL_TEE branch.
• closure upval ref/i64 overwrite cases: skipped due higher setup cost and lower confidence for a smallest regression pass.