215 make memory a parameter

[JavierCladellas]

• Closes make memory a parameter #215

Major changes:

• App config now has a resources field, containing tasks, nodes, exclusive_access and memory values. Conceived to be used along with parameters, like this:

     "resources":{
         "tasks":"{{parameters.resources.tasks.value}}",
         "exclusive_access":"{{parameters.resources.exclusive_access.value}}",
         "memory":"{{parameters.memory.value}}",
     },

• Add a conditions field to parameters, enabling to select given combinations (pruning) of the parameter space.
  For example,

"parameters": [
        {
            "name": "tasks",
             "sequence":[64,128,256]
        },
        {
            "name": "mesh",
             "sequence":["M1","M2","M3","M4"],
            "conditions":{
                "M2":{ "tasks": [128] },
                "M3":{"tasks": [128,256]},
                "M3":{"tasks": []}
            }
        }
]

Will only run the tests:

• M1 with tasks 64, 128, 256 (default value "all" as not specified in conditions)

• M2 only with 128 tasks

• M3 with 128 and 256 tasks

• Will not run any tests for M4
  Also accepts multiple conditions for a given value, e.g "M2":{ "tasks": [128], "discretization": ["P1","P3"] }

• Removed the memory field of the root of the configuration file, it should now be specified under the resources field. The --mem directive is still added to the scheduler script

• [❗ IMPORTANT ❗ ] Removed the use_nodes_option. The --nodes directive will no longer be emitted. But --ntasks and --ntasks-per-node will always appear, we should let the scheduler allocate the number of nodes it needs. ( Done like this because reframe does not let to specify the number of nodes, although we can explicitly pass the --nodes directive but will be scheduler dependent). What do you think @vincentchabannes

• The scheduler script is now exported on the ReFrame report, along with output and error logs. A link to these are now included in the report (adoc) parameter table.

• Separated resource handling resources.py using strategy+factory pattern to handle tasks-nodes-tasks_per_node-memory combinations.

• Also updated tests.

• The actual value of num_nodes, tasks and tasks_per_node | None should be shown in the report, not only the parameter values (as they might differ, be updated )

• Scripts and logs should be exported even on failed tests (if the job fails during the run phase, logs will not be exported 😢 )

• Implement copying mechanism from a input_user_dir folder to input_dataset_base_dir. Only compatible with execution_policy:serial. If input_user_dir is defined, all files listed under input_file_dependencies will be copied (and removed after the test is done).