[Mellanox] Update headroom calculation algorithm

[stephenxs]

What I did

1. Fix a bug in lossless headroom calculation regarding the small packet percentage which should be calculated in a size-based approach instead of packet-based
2. Adjust Mellanox-specific lossless headroom calculation algorithm

Why I did it

How I verified it

Manual test Regression test

Details if related

[stephenxs]

The manual test result is as below. Deviation is small and acceptable port MTU: 9100 RDMA MTU: 1024 Small packet percentage: 50% SPC3

• 400G, 300m.
  • by lua: xon 38912, xoff 272384, size 320512
  • by using internal tool: xon 38k, xoff 265.76k, size 312.8k
• 100G, 40m.
  • by lua: xon 19456, xoff 51200, size 70656
  • by using internal tool: xon 19k, xoff 49.62k, size 68.6k

SPC4

• 800G, 300m.
  • by lua: xon 38912 xoff 474112 size 522240
  • by using internal tool: xon 38k xoff 463.01 size 510.0
• 200G, 40m
  • by lua: xon 19456 xoff 65536 size 84992
  • by using internal tool: xon 19k xoff 63.88k size 84.9k