Performance discrepancy of PingPing vs SendRecv

Using `IntelOneAPI/2025.2` with `IMPI v2021.16` in IMB-MPI1 v2021.10

The official [documentation](https://www.intel.com/content/www/us/en/docs/mpi-library/user-guide-benchmarks/2021-2/sendrecv.html#SENDRECV-DEFINITION) says:

> In the case of two processes, Sendrecv is equivalent to the PingPing benchmark of IMB1.x. For two processes, it reports the bidirectional bandwidth of the system, as obtained by the optimized MPI_Sendrecv function

But when I run it on the exactly same two cores, `sendrecv` outperforms `pingping` for all sizes:

`mpirun -np 2 ./IMB-MPI1 .....`

| #bytes | Mbytes/sec (PingPong) | Mbytes/sec (PingPing) | Mbytes/sec (SendRecv) |
|----------|------------|------------|------------|
| 0 | 0 | 0 | 0 |
| 1 | 4.31 | 3.68 | 7.43 |
| 2 | 8.9 | 7.35 | 14.87 |
| 4 | 16.61 | 14.47 | 28.59 |
| 8 | 32.51 | 29.6 | 58.55 |
| 16 | 68.8 | 53.84 | 118.52 |
| 32 | 141.79 | 115.19 | 238.32 |
| 64 | 255.81 | 181.23 | 373.96 |
| 128 | 500.49 | 391.98 | 807.68 |
| 256 | 927.73 | 715.7 | 1512.49 |
| 512 | 1498.66 | 1203.17 | 2466.14 |
| 1024 | 2532.21 | 1998.36 | 3949.3 |
| 2048 | 4160.18 | 3214.35 | 6291.44 |
| 4096 | 5527.35 | 4414.81 | 9261.44 |
| 8192 | 6522.92 | 5468.8 | 11183.32 |
| 16384 | 7174.1 | 6546.23 | 12956.97 |
| 32768 | 7839.75 | 8239.51 | 17546.81 |
| 65536 | 10287.31 | 8336.17 | 18407 |
| 131072 | 11192.94 | 13060.11 | 26332.25 |
| 262144 | 18100.42 | 13692.36 | 27133.69 |
| 524288 | 19710.77 | 13276.27 | 22186.71 |
| 1048576 | 14577.86 | 8797.04 | 13967.69 |
| 2097152 | 13295.74 | 7716.41 | 15339.54 |
| 4194304 | 13594.43 | 7233.45 | 14542.28 |