Stop treating arm64 sub to a non-sp register as a stack reset ##arch

[phix33]

• Mark this if you consider it ready to merge
• I've added tests (optional)
• I wrote some lines in the book (optional)

Description

On arm64 any sub whose destination is not sp is treated as a stack reset, so after sub x0, x0, 1 the tracked depth drops to 0 and every later sp access in the block is read as a stack argument above the frame.

$ r2 -a arm -b 64 -qc 'e anal.vars.stackname=true; wx ff8300d1000400d1e00300f9ff830091c0035fd6; af; afva; afv; ao 1 @ 4~stackop' -
arg int64_t arg1 @ x0
arg int64_t arg_0h @ sp+0x20
stackop: reset

str x0, [sp] writes the local at the bottom of the 0x20 frame; with this PR it is var_20h @ sp+0x0 and sub x0, x0, 1 has no stackop.

• anop64 no longer sets R_ANAL_STACK_RESET on sub with a non-sp destination, matching add. Only instructions that write sp change the depth.
• arm.v35 had the same reset on both sub and add; it goes too.
• Across the 103 arm64 binaries in the testbins, aaa loses 3587 phantom stack args and gains none.

The new case in test/db/anal/arm64 fails without the fix on both the variable and the stackop; its sub sp, sp, 0x20 line keeps master's inc of 32. A second case checks that arm.v35 no longer reports a reset for sub x0, x0, 1 or add x0, x0, 1; it only runs on builds with v35 enabled.