ASSERT Not implemented core/translate.c:545 walk->xsp_adjust == 0

While trying to implement #1025 (closed), the first run of the decode-stress test hits this assert:

$ bin64/drrun -stress_recreate_state -- suite/tests/bin/common.decode

SYSLOG_ERROR: Application /home/bruening/dr/git/build_x64_dbg_tests/suite/tests/bin/common.decode (131441).  Internal Error: DynamoRIO debug check failure: Not implemented @/home/bruening/dr/git/src/core/translate.c:545 walk->xsp_adjust == 0
(Error occurred @1738 frags)


  0x00007f188a26d000  ff 35 02 90 19 00    push   <rel> 0x00007f188a406008[8byte] %rsp -> %rsp 0xfffffff8(%rsp)[8byte]
  0x00007f188a26d006  ff 25 04 90 19 00    jmp    <rel> 0x00007f188a406010[8byte]

=>
Fragment 1738, tag 0x00007f188a26d000, flags 0x1000018, shared, tracehead, size 83:
        
  0x00007f1845f2cee4  65 48 a3 00 00 00 00 mov    %rax -> %gs:0x00[8byte]
                      00 00 00 00
  0x00007f1845f2ceef  48 b8 08 60 40 8a 18 mov    $0x00007f188a406008 -> %rax
                      7f 00 00
  0x00007f1845f2cef9  ff 30                push   (%rax)[8byte] %rsp -> %rsp 0xfffffff8(%rsp)[8byte]
  0x00007f1845f2cefb  65 48 a1 00 00 00 00 mov    %gs:0x00[8byte] -> %rax
                      00 00 00 00
  0x00007f1845f2cf06  65 48 a3 00 00 00 00 mov    %rax -> %gs:0x00[8byte]
                      00 00 00 00
  0x00007f1845f2cf11  48 b8 10 60 40 8a 18 mov    $0x00007f188a406010 -> %rax
                      7f 00 00
  0x00007f1845f2cf1b  65 48 89 0c 25 10 00 mov    %rcx -> %gs:0x10[8byte]
                      00 00
  0x00007f1845f2cf24  48 8b 08             mov    (%rax)[8byte] -> %rcx
  0x00007f1845f2cf27  65 48 a1 00 00 00 00 mov    %gs:0x00[8byte] -> %rax
                      00 00 00 00
  0x00007f1845f2cf32  e9 09 ba f8 ff       jmp    $0x00007f1845eb8940 <shared_bb_ibl_indjmp>

  
ilist for recreation:
TAG  0x00007f188a26d000
 +0    m4 @0x00007f1645f5bab8  65 48 a3 00 00 00 00 mov    %rax -> %gs:0x00[8byte]
                               00 00 00 00
 +11   m4 @0x00007f1645f5f280  48 b8 08 60 40 8a 18 mov    $0x00007f188a406008 -> %rax
                               7f 00 00
 +21   L3 @0x00007f1645f87498  ff 30                push   (%rax)[8byte] %rsp -> %rsp 0xfffffff8(%rsp)[8byte]
 +23   m4 @0x00007f1645f5e4a8  65 48 a1 00 00 00 00 mov    %gs:0x00[8byte] -> %rax
                               00 00 00 00
 +34   m4 @0x00007f1645f5dc58  65 48 a3 00 00 00 00 mov    %rax -> %gs:0x00[8byte]
                               00 00 00 00
 +45   m4 @0x00007f1645f5b1c8  48 b8 10 60 40 8a 18 mov    $0x00007f188a406010 -> %rax
                               7f 00 00
 +55   m4 @0x00007f1645f5fa68  65 48 89 0c 25 10 00 mov    %rcx -> %gs:0x10[8byte]
                               00 00
 +64   L3 @0x00007f1645f5f768  48 8b 08             mov    (%rax)[8byte] -> %rcx
 +67   m4 @0x00007f1645f5ae00  65 48 a1 00 00 00 00 mov    %gs:0x00[8byte] -> %rax
                               00 00 00 00
 +78   L4 @0x00007f1645f5ac20  e9 ab 29 f6 ff       jmp    $0x00007f1845eb8940 <shared_bb_ibl_indjmp>
END 0x00007f188a26d000

recreate_app : looking for 0x00007f1845f2cefb in frag @ 0x00007f1845f2cee4 (tag 0x00007f188a26d000)
        ok instr: mov    %rax -> %gs:0x00[8byte]
translate_walk_track: entering mangle region xl8=0x00007f188a26d000
        spill update: spill tls rax offs=0
        ok instr: mov    $0x00007f188a406008 -> %rax
        ok instr: push   (%rax)[8byte] %rsp -> %rsp 0xfffffff8(%rsp)[8byte]
        state track: push or push_imm
recreate_app -- found valid state pc 0x00007f188a26d006
        translation 0x00007f188a26d006 is in mangling epilogue 0x00007f188a26d000 checking for simple symmetric mangling case
SYSLOG_ERROR: Application /home/bruening/dr/git/build_x64_dbg_tests/suite/tests/bin/common.decode (131441).  Internal Error: DynamoRIO debug check failure: Not implemented @/home/bruening/dr/git/src/core/translate.c:545 walk->xsp_adjust == 0

mangle_rel_addr() has the push as a mangle instr:

            /* we need the whole spill...restore region to all be marked mangle */
            instr_set_our_mangling(instr, true);

And see the big comment before the call to instr_supports_simple_mangling_epilogue() where it sets the restore to an epilogue for non-cti: that's from #3307, PR #3318.

So do we relax the assert, or do we include push/pop with cti's for not setting to epilogue and thus xl8 to pre-app-instr PC and re-execute the push/pop after adjusting xsp?? Leaning toward former since unlike cti we can easily xl8 to post-app-instr PC and it seems safer to not re-do the push.