Rm legacy essentia analysis

[ffont]

Description
Remove usage of legacy Essentia extractor. We've been running "legacy" and "v1" extractors for a while. They are mostly compatible although there are some small differences. We need to switch to "v1" for maintainability, etc.

Main differences:

• v1 does not include chords_count and chords_progression descriptors: this is something we decided years ago, they don't make sense for most of the sounds and also they are not very reliable (although this should be properly evaluated). We could re-incorporate them in the future, but I would say not without a proper evaluation.
• In v1 we only analyze first 60 seconds of sounds (80% of fs is less than 60s long). This is to make the analyzer faster
• In v1 we don't save frame-by-frame analysis results. This means we save a lot of disk space. People is not accessing these files. Also this detailed analysis feature is well suited for local audio analysis.

Deployment steps:
After this is merged, we need to update local settings of deployment configuration so that we no longer use legacy analyzer for any similarity space and we remove it from analyzer's configuration dict. Then we can get rid of all SoundAnalysis objects that belong to the legacy analyzer, and also all files on disk (including frames analysis). We should re-run "consolidate analysis" for all files so descriptor values are now loaded from the new extractor (this is is actually not mandatory, but nice to do in case there were some differences between legacy and v1).