Improve deviceSharing configuration flexibility for heterogeneous environments

Currently MortalGPU requires to define metagpusPerGpu static amount in the sharing configuration. Moreover the sharing configuration mapping to GPU devices is only possible by UUIDs, that is not flexible in dynamic cloud environment.

There is a valid use-case, especially for cloud-provisioned infra, where nodes has different GPU models with different amount of memory (see #10 (closed)).

Proposed sharing configuration logic improvement:

• in GPU discovery loop, when device is matched against the sharing configuration, define number of meta-devices based on the defined allocatable memory chunk size (calculate automatically based on GPU memory)
• introduce filtering by GPU model, defined by mortals as string, in addition to UUIDs