Faster large collections

[SmilyOrg]

Fixes a few smaller and bigger issues when it comes to performance loading 100k+ file collections.

Improvements

Tests with collections made from 10M+ dummy files reveal the following improvements. Numbers based on an older i7-5820K 6-Core desktop CPU.

• 10M files in 1 dir
  • Before: loading starts after 15 seconds
  • After: loading starts in under 1 second
• 10M files in 3000 dirs
  • Before: loading starts after 38 seconds
  • After: loading starts in under 1 second

Changes

• Fix the loading spinner only showing the number without "files"
• Add skip_collection_counts to avoid showing collection counts on startup, which can speed it up in case of large collections
• Refactor the way collection loading works to avoid some slow operations
  • Reduce the amount of objects created leading to less garbage having to be collected
  • Avoid SQLite sorting the collection in memory, leading to faster big collection loads
  • Fix large collections breaking transformation matrix inversion, fixes panic: determinant of affine transformation matrix is zero #54
• Fixed a bug where collections with dirs that are a prefix of other dirs would in some cases erroneously also list files in these other dirs (e.g. /vacation would include /vacation2 or /vacation-eu)

Limitations / Breaking changes

This change introduces a limitation of max 50000 directories that can be sourced in a single collection. Considering the previously long loading time of such collections, this is probably not a big concern. If you hit this limitation, please let me know!