[355_wip] Let dynamo capture silu+mul+f4gemm pattern

[xytpai]

we registered a new pattern-matching logic for integrating aiter::act_mul_and_mxfp4_quant

[xytpai]

BASELINE:

local-completions (model=/data/Llama-3.3-70B-Instruct-MXFP4-Preview,base_url=http://localhost:9771/v1/completions,num_concurrent=256,max_retries=10,max_gen_toks=2048), gen_kwargs: (None), limit: 250.0, num_fewshot: 5, batch_size: auto
|Tasks|Version|     Filter     |n-shot|  Metric   |   |Value|   |Stderr|
|-----|------:|----------------|-----:|-----------|---|----:|---|-----:|
|gsm8k|      3|flexible-extract|     5|exact_match|↑  |0.932|±  |0.0160|
|     |       |strict-match    |     5|exact_match|↑  |0.056|±  |0.0146|
============ Serving Benchmark Result ============
Successful requests:                     256       
Maximum request concurrency:             64        
Request rate configured (RPS):           2.00      
Benchmark duration (s):                  290.21    
Total input tokens:                      261888    
Total generated tokens:                  262144    
Request throughput (req/s):              0.88      
Output token throughput (tok/s):         903.28    
Total Token throughput (tok/s):          1805.68   
---------------Time to First Token----------------
Mean TTFT (ms):                          2107.39   
Median TTFT (ms):                        1916.77   
P99 TTFT (ms):                           4252.91   
-----Time per Output Token (excl. 1st token)------
Mean TPOT (ms):                          65.00     
Median TPOT (ms):                        67.95     
P99 TPOT (ms):                           69.83     
---------------Inter-token Latency----------------
Mean ITL (ms):                           65.00     
Median ITL (ms):                         42.93     
P99 ITL (ms):                            988.55    
----------------End-to-end Latency----------------
Mean E2EL (ms):                          68598.03  
Median E2EL (ms):                        72068.74  
P99 E2EL (ms):                           72315.47  
==================================================

NEW:

local-completions (model=/data/Llama-3.3-70B-Instruct-MXFP4-Preview,base_url=http://localhost:9771/v1/completions,num_concurrent=256,max_retries=10,max_gen_toks=2048), gen_kwargs: (None), limit: 250.0, num_fewshot: 5, batch_size: auto
|Tasks|Version|     Filter     |n-shot|  Metric   |   |Value|   |Stderr|
|-----|------:|----------------|-----:|-----------|---|----:|---|-----:|
|gsm8k|      3|flexible-extract|     5|exact_match|↑  |0.952|±  |0.0135|
|     |       |strict-match    |     5|exact_match|↑  |0.052|±  |0.0141|

============ Serving Benchmark Result ============
Successful requests:                     256       
Maximum request concurrency:             64        
Request rate configured (RPS):           2.00      
Benchmark duration (s):                  282.66    
Total input tokens:                      261888    
Total generated tokens:                  262144    
Request throughput (req/s):              0.91      
Output token throughput (tok/s):         927.42    
Total Token throughput (tok/s):          1853.93   
---------------Time to First Token----------------
Mean TTFT (ms):                          2229.05   
Median TTFT (ms):                        2081.75   
P99 TTFT (ms):                           4068.82   
-----Time per Output Token (excl. 1st token)------
Mean TPOT (ms):                          63.08     
Median TPOT (ms):                        65.72     
P99 TPOT (ms):                           68.25     
---------------Inter-token Latency----------------
Mean ITL (ms):                           63.08     
Median ITL (ms):                         42.06     
P99 ITL (ms):                            937.36    
----------------End-to-end Latency----------------
Mean E2EL (ms):                          66760.35  
Median E2EL (ms):                        69354.50  
P99 E2EL (ms):                           70977.04  
==================================================

[xytpai]

[xytpai]

To enable mxfp4 fusion:

--compilation-config '{"cudagraph_mode": "FULL_AND_PIECEWISE", "custom_ops": ["+rms_norm", "+silu_and_mul"]}'

[xytpai]

To reduce the number of PRs, this PR has been merged into #705