[TTS] Add dataset to path of logged artifacts

[rlangman]

What does this PR do ?

Merging #7462 to r1.21.0

Modify callbacks to save artifacts into dataset specific folders, instead of saving artifacts from all datasets into 1 folder. This helps organization when training on many datasets, and also handles the case when multiple datasets have files with the same name.

Collection: [TTS]

Changelog

• Pass dataset names through TTS data loaders
• Save artifacts under epoch_{x}/{dataset_name}/
• Add initial logging for FastPitch and HiFi-GAN recipes

Before your PR is "Ready for review"

Pre checks:

• Make sure you read and followed Contributor guidelines
• Did you write any new necessary tests?
• Did you add or update any necessary documentation?
• Does the PR affect components that are optional to install? (Ex: Numba, Pynini, Apex etc)
  • Reviewer: Does the PR have correct import guards for all optional libraries?

PR Type:

• New Feature
• Bugfix
• Documentation