Logger connector improvements tracker

[carmocca]

🚀 Feature

• Logger connector read-only properties: https://github.com/PyTorchLightning/pytorch-lightning/pull/7882/files#r647359370
• test fx_validator for all functions: https://github.com/PyTorchLightning/pytorch-lightning/blob/764d2c775e2f3568975a60e94520bca8a42b2490/pytorch_lightning/trainer/connectors/logger_connector/fx_validator.py#L83. Always use trainer.call_hook #8498
• Refactor logging tests - they are currently scattered over several files.
• Rely on Metric to sync. We gather+reduce together (per training type) whereas torchmetrics does it separately. Would also need to inject is_tpu_distributed() logic. https://github.com/PyTorchLightning/pytorch-lightning/blob/764d2c775e2f3568975a60e94520bca8a42b2490/pytorch_lightning/trainer/connectors/logger_connector/result_new.py#L176-L193

Fault Tolerant Logging

• Serialization: bab742e 37e7589
• Add load_from_state_dict to ResultCollection
• Design loop state save/load hierarchy. WIP in 189f0ad. Intended structure:

trainer
    loops:
        fit_loop:
            epoch_loop:
                batch_loop:
                validation_loop:
                    ...
        validate_loop:
            ...
        test_loop:
            ...
------
and each loop has:
a_loop:
    progress
    results

• Include the loops state in the checkpoint