prov/efa, fabtests: Fix cq data size in efa-rdm pkt post, fix cq_data test bugs

[shijin-aws]

commit 10ca04b introduced a bug that uses a
16 bit cq data intermediate variable when passing to
device. uint16_t is too small and
will silently corrupt when application is using more
than 16 bits while efa provider supports 32 bits.
This patch fixes it.

[shijin-aws]

@j-xiong can u review the fabtests change

[a-szegel]

Hey @j-xiong, Why did this PR fail intel CI?

[shijin-aws]

the appveyor failure looks unrelated

 name:   rdm_atomic -o all -p tcp
  timestamp: Tue 11/25/2025 17:17:47.13
  result: Fail
  time:   13
  server_cmd: C:\projects\libfabric\fabtests\x64\Debug-v142\rdm_atomic -o all -p tcp -I 10 -s 127.0.0.1
  server_stdout:
hmem allocation error: Not enough space
  client_cmd: C:\projects\libfabric\fabtests\x64\Debug-v142\rdm_atomic -o all -p tcp -I 10 -s 127.0.0.1 127.0.0.1
  client_stdout:
hmem allocation error: Not enough space
- name:   rdm_atomic -o all -v -p tcp
  timestamp: Tue 11/25/2025 17:17:57.13
  result: Fail
  time:   10
  server_cmd: C:\projects\libfabric\fabtests\x64\Debug-v142\rdm_atomic -o all -v -p tcp -I 10 -s 127.0.0.1
  server_stdout:
hmem allocation error: Not enough space
  client_cmd: C:\projects\libfabric\fabtests\x64\Debug-v142\rdm_atomic -o all -v -p tcp -I 10 -s 127.0.0.1 127.0.0.1
  client_stdout:
hmem allocation error: Not enough space
- name:   rdm_cntr_pingpong -p tcp

[aingerson]

@a-szegel Unrelated mpich test suite failure with tcp - bad file descriptor. You can ignore