Feature: support putmem nbi with signal && delete bnxt init extra malloc

[jhchouuu]

Motivation

support putmem nbi with signal

Technical Details

• will allocate all the necessary wq slots for both write wqe and signal wqe at one time, to achieve a compact arrival between "write" and "signal" to the remote endpoint.
• template "onlyOneSignal" is used to control whether all threads in a warp send the signal or only one thread sends the signal.
• delete bnxt init extra malloc.

test example

test command:
MORI_DISABLE_P2P=1 MORI_GLOBAL_LOG_LEVEL=info mpirun --allow-run-as-root -np 2 ./examples/concurrent_put_signal_thread