ASSERT "internal isa mode error" on ubuntu22 in many tests

[derekbruening]

For #7270 we need to move our tests to Ubuntu22.
Many fail there with this assert, here in client.tls:

in foo_t::foo_t
<curiosity: rex.w on OPSZ_6_irex10_short4!>
<Application /home/derek/dr/git/build_x64_dbg_tests/suite/tests/bin/client.tls (169991) DynamoRIO usage error : internal isa mode error>
<Usage error: internal isa mode error (/home/derek/dr/git/src/core/ir/decode_shared.c, line 261)

The assert is here:

#0  report_dynamorio_problem (dcontext=0x0, dumpcore_flag=16, exception_addr=0x0, report_ebp=0x0, fmt=0x7ffff7d71e57 "Usage error: %s (%s, line %d)")
    at /home/derek/dr/git/src/core/utils.c:2124
#1  0x00007ffff7aee10b in external_error (file=0x7ffff7dad588 "/home/derek/dr/git/src/core/ir/decode_shared.c", line=261, msg=0x7ffff7dad5bd "internal isa mode error")
    at /home/derek/dr/git/src/core/utils.c:201
#2  0x00007ffff7c5ff9e in dr_set_isa_mode (drcontext=0x0, new_mode=DR_ISA_AMD64, old_mode_out=0x7ffdb37e8914) at /home/derek/dr/git/src/core/ir/decode_shared.c:261
#3  0x00007ffff7c60157 in set_x86_mode (drcontext=0x0, x86=false) at /home/derek/dr/git/src/core/ir/x86/decode.c:150
#4  0x00007ffff7c639b4 in read_instruction (pc=0x7ffff7297544 "\213W\020\270\026", orig_pc=0x7ffff7297540 "\363\017\036\372\213W\020\270\026", ret_info=0x7ffdb37e8c58,
    di=0x7ffdb37e8bd0, just_opcode=false, report_invalid=true) at /home/derek/dr/git/src/core/ir/x86/decode.c:1328
#5  0x00007ffff7c67501 in decode_common (dcontext=0xffffffffffffffff, pc=0x7ffff7297540 "\363\017\036\372\213W\020\270\026", orig_pc=0x7ffff7297540 "\363\017\036\372\213W\020\270\026",
    instr=0x7ffdb37e8cf0) at /home/derek/dr/git/src/core/ir/x86/decode.c:2619
#6  0x00007ffff7c68052 in decode (drcontext=0xffffffffffffffff, pc=0x7ffff7297540 "\363\017\036\372\213W\020\270\026", instr=0x7ffdb37e8cf0)
    at /home/derek/dr/git/src/core/ir/x86/decode.c:2782
#7  0x00007ffff7d25610 in privload_set_pthread_tls_fields (mod=0x7ffdb36e9478, priv_tls_base=0x7ffdb37f0000 "") at /home/derek/dr/git/src/core/unix/loader_linux.c:321
#8  0x00007ffff7d25f50 in privload_copy_tls_block (mod=0x7ffdb36e9478, priv_tls_base=0x7ffdb37f0000 "", mod_idx=0) at /home/derek/dr/git/src/core/unix/loader_linux.c:413
#9  0x00007ffff7d2637e in privload_tls_init (app_tp=0x7ffff7000640) at /home/derek/dr/git/src/core/unix/loader_linux.c:509
#10 0x00007ffff7ce6b1c in os_tls_app_seg_init (os_tls=0x7ffdb37ec000, segment=0x7ffdb37ec000) at /home/derek/dr/git/src/core/unix/os.c:2296
#11 0x00007ffff7ce6ec0 in os_tls_init () at /home/derek/dr/git/src/core/unix/os.c:2362
#12 0x00007ffff7a63a94 in dynamo_thread_init (dstack_in=0x7ffdb37ea000 "", mc=0x7ffdb37e8f80, os_data=0x7ffdb37e9880, client_thread=false) at /home/derek/dr/git/src/core/dynamo.c:2302
#13 0x00007ffff7cc9a56 in new_thread_setup (mc=0x7ffdb37e8f80) at /home/derek/dr/git/src/core/arch/x86_code.c:305

That's the #6611 pthread init issue.
I put in the TLS init code right before dcontext is in the thread slot.
It apparently is the only post-global-init-pre-thread-init
decode done anywhere.
It does use GLOBAL_DCONTEXT but the read_instruction call loses that. So I
will either pass in dcontext to read_instruction which seems cleaner but
adds work on the decode path that's not needed for release build...or relax
the assert.