Make max_num_batched_tokens behavior more verbose, add legacy mode

[kzawora-intel]

Addressing issues from #207
Now, filtering behavior is more verbose, handling common errors and displaying numbers of omitted buckets due to token budget (in debug log level, buckets are printed):

INFO 08-27 20:57:27 profiler.py:62] Profiler enabled for: vllm-instance-1ab4f6c4d726480d8825044cf74e9af1
WARNING 08-27 20:57:27 utils.py:566] Pin memory is not supported on HPU.
INFO 08-27 20:57:27 selector.py:85] Using HabanaAttention backend.
INFO 08-27 20:57:27 habana_model_runner.py:563] Prompt bucket config (min, step, max_warmup) bs:[1, 32, 64], seq:[128, 128, 1024]
INFO 08-27 20:57:27 habana_model_runner.py:576] Generated 23 prompt buckets: [(1, 128), (1, 256), (1, 384), (1, 512), (1, 640), (1, 768), (1, 896), (1, 1024), (2, 128), (2, 256), (2, 384), (2, 512), (2, 640), (2, 768), (2, 896), (2, 1024), (4, 128), (4, 256), (4, 384), (4, 512), (8, 128), (8, 256), (16, 128)]
INFO 08-27 20:57:27 habana_model_runner.py:581] Omitted 33 prompt buckets due to exceeded token budget (max_num_batched_tokens=2048)
INFO 08-27 20:57:27 habana_model_runner.py:589] Decode bucket config (min, step, max_warmup) bs:[1, 128, 256], seq:[128, 128, 2048]
INFO 08-27 20:57:27 habana_model_runner.py:600] Generated 31 decode buckets: [(1, 128), (1, 256), (1, 384), (1, 512), (1, 640), (1, 768), (1, 896), (1, 1024), (1, 1152), (1, 1280), (1, 1408), (1, 1536), (1, 1664), (1, 1792), (1, 1920), (1, 2048), (2, 128), (2, 256), (2, 384), (2, 512), (2, 640), (2, 768), (2, 896), (2, 1024), (4, 128), (4, 256), (4, 384), (4, 512), (8, 128), (8, 256), (16, 128)]
INFO 08-27 20:57:27 habana_model_runner.py:605] Omitted 113 decode buckets due to exceeded token budget (max_num_batched_tokens=2048)

Legacy mode was also added, which throws a nasty error message whenever token budget is set too low, but then it omits filtering and works as it did previously (ran with VLLM_DECODE_BS_BUCKET_MIN=128 VLLM_DECODE_SEQ_BUCKET_MIN=1024 python vllm_test.py --max-num-batched-tokens=2048):

INFO 08-27 21:01:02 profiler.py:62] Profiler enabled for: vllm-instance-51f60d3978d347e992436f1dc0aa4702
WARNING 08-27 21:01:02 utils.py:566] Pin memory is not supported on HPU.
INFO 08-27 21:01:02 selector.py:85] Using HabanaAttention backend.
INFO 08-27 21:01:02 habana_model_runner.py:563] Prompt bucket config (min, step, max_warmup) bs:[1, 32, 64], seq:[128, 128, 1024]
INFO 08-27 21:01:02 habana_model_runner.py:576] Generated 23 prompt buckets: [(1, 128), (1, 256), (1, 384), (1, 512), (1, 640), (1, 768), (1, 896), (1, 1024), (2, 128), (2, 256), (2, 384), (2, 512), (2, 640), (2, 768), (2, 896), (2, 1024), (4, 128), (4, 256), (4, 384), (4, 512), (8, 128), (8, 256), (16, 128)]
INFO 08-27 21:01:02 habana_model_runner.py:581] Omitted 33 prompt buckets due to exceeded token budget (max_num_batched_tokens=2048)
INFO 08-27 21:01:02 habana_model_runner.py:589] Decode bucket config (min, step, max_warmup) bs:[128, 128, 256], seq:[1024, 128, 2048]
ERROR 08-27 21:01:02 habana_model_runner.py:128] The current bucketing configuration (min, step, max_warmup): bs:[128, 128, 256], seq:[1024, 128, 2048] cannot be used with specified max_num_batched_tokens (2048), as the smallest bucket (16384) would exceed token budget. Please increase max_num_batched_tokens or decrease bucket minimum Ignoring max_num_batched_tokens at risk of out-of-memory errors.
INFO 08-27 21:01:02 habana_model_runner.py:600] Generated 32 decode buckets: [(128, 128), (128, 256), (128, 384), (128, 512), (128, 640), (128, 768), (128, 896), (128, 1024), (128, 1152), (128, 1280), (128, 1408), (128, 1536), (128, 1664), (128, 1792), (128, 1920), (128, 2048), (256, 128), (256, 256), (256, 384), (256, 512), (256, 640), (256, 768), (256, 896), (256, 1024), (256, 1152), (256, 1280), (256, 1408), (256, 1536), (256, 1664), (256, 1792), (256, 1920), (256, 2048)]
INFO 08-27 21:01:02 habana_model_runner.py:605] Omitted 0 decode buckets due to exceeded token budget (max_num_batched_tokens=2048)

[kdamaszk]

LGTM

[szutenberg]

It fixes my workloads, thanks!