Parallelism fixes

[kalantar]

Fixes and enhances support for specifying parallelism. In particular, it:

• adds parallelism.dataLocal and parallelism.workers to decode and prefill
• it updates the calculation of each value taking into account the relationships between them and reports problems if the relationships are invalid
• it adds the --data-parallel-size vllm option (superceding Add --data-parallel-size to args if DP_SIZE gt 1 #126)
• it adds the --data-parallel-size-local vllm option
• it defines a DP_SIZE_LOCAL environment variable in the prefill and decode vllm containers
• it corrects the computation of the number of gpus per worker (superceding If user set acceleratorResource, respect the value #141 and fixes acceleratorResource is enforced to be equal to tensorParallelism #140)
• it corrects the computation of the number of workers (size) in a LeaderWorkerSet
• updates README.md

In addition, this PR includes the following update to the examples:

• updates the make generate command to include all examples
• added accelerator.type to identify the type of (gpu) resources to request instead of having it explicit (and sometimes wrong) in the resource request