[GR-18266] Add UseKill annotation to replace the Use+Temp trick.

[graalvmbot]

This PR introduces the new @UseKill operand annotation to the LIR and Graal compiler codebase, replacing the previous practice of using @Use+@Temp (the "Use+Temp trick") for denoting operands that are both read and then killed (i.e., whose register may be reused after the instruction). The change improves register allocation logic by making operand liveness semantics explicit and easier to reason about, and extends support throughout register allocation, stack slot allocation, verification, and various architecture-specific LIR instructions.

Key Changes:

• Added the @UseKill annotation and OperandMode.USE_KILL to LIRInstruction, including new API methods (forEachUseKill, visitEachUseKill) and related wiring in LIRInstructionClass.
• Implemented the UseKillMoveInjectionPhase, a pre-allocation LIR phase that lowers @UseKill operands on variables to explicit moves, ensuring correct semantics for variables marked as killed.
• Replaced all manual use of @Use+@Temp with the new @UseKill, including in key LIR instructions such as AMD64ArrayEqualsOp, AMD64VectorizedMismatchOp, and AMD64ArrayRegionCompareToOp.
• Updated register allocation (LinearScanLifetimeAnalysisPhase, LinearScanAssignLocationsPhase, RegisterVerifier), stack slot allocation (LSStackSlotAllocator, SimpleStackSlotAllocator, FixPointIntervalBuilder), economy and full pre-allocation LIR phases to include the new lowering phase.
• Modified verification, codegen, optimization, and marking/logging utilities (SSAVerifier, LocationMarker, HotSpotZapRegistersPhase, etc.) to consistently handle USE_KILL wherever operand modes are enumerated.
• Backward-compatible handling of cast values (e.g., SimpleVirtualStackSlotAlias) with USE_KILL mode checks.
• Extended operand walking throughout all relevant code locations to include visitEachUseKill and forEachUseKill.