offsetB assert in genSpill_ARM64

### Context

Running a large program results in a valgrind assert

#### What went wrong?

When running the program it hits the following assert:
vex: priv/host_arm64_defs.c:2836 (void genSpill_ARM64(HInstr **, HInstr **, HReg, Int, Bool)): Assertion `offsetB < 4096' failed.
Command run is just plain "valgrind".

To my best understanding, while the spill area itself is only 4096 bytes there is an additional offset involved, meaning offsetB can become larger than 4096.
Best I can tell the simplest fix should be to change ARM64Instr_VLdStQ to use ldr/str instead of ld1.2d/st1.2d, since the former supports a 12-bit offset.
This both avoids the need for the extra add in HRcVec128 and increases the range to 64k.

### Information

* macOS architecture (`uname -m`): arm64
* macOS version (`sw_vers`): 26.5.1
* Xcode version (`xcrun --sdk macosx --show-sdk-version`): 26.5

Start in priv/host_arm64_defs.c at genSpill_ARM64 and inspect ARM64Instr_VLdStQ and HRcVec128, then reproduce the assertion with the reported large program under plain valgrind. Compare the offset handling with the proposed ldr/str approach. Done means the ARM64 run no longer triggers the offsetB assertion for the reported case.

Written by the indexing model from the issue text.