q and ge flags are not considered "arithmetic" in defines, yet are in flag preservation code

Xref recent CDash failures on ARM (e.g., http://dynamorio.org/CDash/testDetails.php?test=243665&build=21890 after I added drreg_restore_app_aflags() tests to drreg-test in 4afa0bd0.

The problem first shows up on OP_sel:

interp: start_pc = 0xb6f54730
  0xb6f54730  fa82 f24c  uadd8  %r2 %r12 -> %r2
  0xb6f54734  faa4 f28c  sel    %r4 %r12 -> %r2
  0xb6f54738  fa83 f34c  uadd8  %r3 %r12 -> %r3
  0xb6f5473c  faa2 f38c  sel    %r2 %r12 -> %r3
  0xb6f54740  bb4b       cbnz   $0xb6f54796 %r3
end_pc = 0xb6f54742

after instrumentation:
TAG  0xb6f54730
 +0    m4 @0x469a8a7c  f8ca 0084  str    %r0 -> +0x84(%r10)[4byte]
 +4    m4 @0x469a682c  f3ef 8000  mrs    %cpsr -> %r0
 +8    m4 @0x469a8244  f8ca 0080  str    %r0 -> +0x80(%r10)[4byte]
 +12   m4 @0x469a8e4c  f8da 0080  ldr    +0x80(%r10)[4byte] -> %r0
 +16   m4 @0x469a6748  f380 8c00  msr    $0x0c %r0 -> %cpsr
 +20   L3              fa82 f24c  uadd8  %r2 %r12 -> %r2
 +24   m4 @0x469a8a30  f8da 0080  ldr    +0x80(%r10)[4byte] -> %r0
 +28   m4 @0x469a84c8  f380 8c00  msr    $0x0c %r0 -> %cpsr
 +32   L3              faa4 f28c  sel    %r4 %r12 -> %r2
 +36   L3              fa83 f34c  uadd8  %r3 %r12 -> %r3
 +40   L3              faa2 f38c  sel    %r2 %r12 -> %r3
 +44   m4 @0x469a89e4  f8da 0080  ldr    +0x80(%r10)[4byte] -> %r0
 +48   m4 @0x469a8c44  f380 8c00  msr    $0x0c %r0 -> %cpsr
 +52   m4 @0x469a81ac  f8da 0084  ldr    +0x84(%r10)[4byte] -> %r0
 +56   L3              bb4b       cbnz   $0xb6f54796 %r3
END 0xb6f54730

    {OP_uadd8,    0xfa80f040, "uadd8",  RCw, xx, RAw, RDw, xx, no, fWGE, END_LIST},
    {OP_sel,      0xfaa0f080, "sel",    RCw, xx, RAw, RDw, xx, no, fRGE, END_LIST},

Looks like we don't update the stored app aflags after the app write OP_uadd8.

It's b/c DR considers only nzcv to be the "arithmetic" flags on ARM in EFLAGS_WRITE_ARITH. Yet dr_restore_arith_flags_from_reg() uses apsr_nzcvqg, so we're clobbering GE.

Unless someone remembers a reason to not include qg in "arith", that seems to be the fix. My notes don't seem to have an explicit decision to exclude them.