Paged attention support for multi gpu

[EricLBuehler]

• Maximum PA. GPU allocation is now equivalent of the model max. seq length (should this not be included for the case of multiple requests?)
• Handle different GPUs with different memory availability and therefore a lower max kv cache size for the model
• Better measure memory on Metal devices

[github-actions]

Code Metrics Report

  ===============================================================================
 Language            Files        Lines         Code     Comments       Blanks
===============================================================================
 C Header                2           35           28            0            7
 Dockerfile              1           41           22           10            9
 JSON                   12          105          104            0            1
 Python                 64         2729         2359           71          299
 Shell                   1           57           22           18           17
 Plain Text              3         3723            0         2413         1310
 TOML                   18          611          544            2           65
 YAML                    2           21           19            2            0
-------------------------------------------------------------------------------
 Jupyter Notebooks       4            0            0            0            0
 |- Markdown             2           77           32           31           14
 |- Python               2          205          178            1           26
 (Total)                            282          210           32           40
-------------------------------------------------------------------------------
 Markdown               44         3460            0         2625          835
 |- BASH                 6          103          100            0            3
 |- JSON                 1           12           12            0            0
 |- Python               7          121          109            0           12
 |- Rust                13          440          373            0           67
 |- TOML                 2           75           63            0           12
 (Total)                           4211          657         2625          929
-------------------------------------------------------------------------------
 Rust                  299        94304        84616         1921         7767
 |- Markdown           145         1617           25         1469          123
 (Total)                          95921        84641         3390         7890
===============================================================================
 Total                 450       105086        87714         7062        10310
===============================================================================