[Pipe CLI] Allow to launch runs on capacity blocks via `pipe`

[ekazachkova]

Background
At the moment, Cloud Pipeline does not support launching runs on capacity blocks and it would be nice to support it.

Approach

• retrieve instance type (directly from input arguments or fetch from API)
• fetch launch.reservation.parameters preference:

# Structure description:
{ 
 "<instance type>": {
  "tag": "<name>",
  "gpu_requests_enabled": true/false,
  "cpu_requests_enabled": true/false,
  "ram_requests_enabled": true/false,
  "parameters": {
   "<parameter name>": "<parameter value>", ...
  },
  "kube_assign_policy": { ... }
 }, ...
}

• if instance type contained in launch.reservation.parameters process such runs on capacity block:
  • if gpu_requests_enabled/cpu_requests_enabled/ram_requests_enabled enabled in launch.reservation.parameters (for requested instance type) the corresponding parameters CP_CAP_REQUESTS_GPU/CP_CAP_REQUESTS_CPU/CP_CAP_REQUESTS_RAM shall be specified for this run.
  • parameters content shall be added to run parameters (input arguments can override them).
  • pass kube_assign_policy to run request object (podAssignPolicy field)
• if instance type not matches launch.reservation.parameters pipe run command shall proceed as now.