v2.1.2

• feat(s3): add a use_threads config (#33)

v2.1.1

• feat: allow pillow 13 (#32)

v2.1.0

• feat: add webp to images (#29)
• fix(local): anchor relative FS_ROOT to CWD for all code paths (#30)
• fix(s3): store SVG files as attachments, so opening one never runs its scripts (#31)

v2.0.3

• feat(s3): only create missing buckets when CREATE_BUCKET is set, so startup does not depend on S3 (#28)

v2.0.2

• fix: give compressed files the archive mime type instead of their content one (#27)

v2.0.1

• ci: do not run a second pipeline on tags, which raced the branch one to publish (#26)
• fix(s3): upload without spawning a thread pool per transfer (#25)

v2.0.0

• feat(s3)!: stream uploads with upload_fileobj for S3 instead of a single PUT and remove fake checksum returned (#24)
  save() used to read the whole file in memory before a single PUT, so
  an upload was bounded by the RAM of the worker and by the 5GB limit of a
  PUT. It now goes through upload_fileobj, which reads the file by
  blocks and switches to a multipart upload past 8MB.

  It also asks S3 to checksum what it receives, so a corrupted transfer is
  rejected and metadata() still reports a digest of the content whatever
  the number of parts it travelled in. It's a CRC32, the only algorithm S3
  combines over a whole object. checksum is None only for objects
  written before this.

  Checked against the OVH bucket: a 12MB upload comes back with
  ChecksumType: FULL_OBJECT and its CRC32, and a wrong digest is
  rejected with BadDigest.

v1.4.6

• fix: do not deep copy the storage when mongoengine copies a FileField (#23)

v1.4.5

• chore: apply ruff (#22)
• chore: switch to uv and tag version script (#21)
• feat: allow setting a prefix for all files (#20)
• fix: delete filename always by prefix in S3 (#19)
• fix(s3): fix content type if not guessed (#18)
• Update to version 1.4.5.dev for next development cycle