Add xattr-based attribute support to LocalFileSystem

[Barre]

Closes #331

Store object attributes as extended attributes on the local filesystem, enabling put_opts and put_multipart_opts to persist attributes. Attributes are set on the staging file before atomic rename to preserve consistency.

• Add xattr dependency (gated by fs feature)
• Map standard attributes to user.* xattr namespace
• Read attributes back in get_opts

Rationale for this change

The LocalFileSystem currently does not support attributes.

What changes are included in this PR?

The LocalFileSystem now supports attributes.

Are there any user-facing changes?

The LocalFileSystem now supports attributes.

[crepererum]

I think this needs fixes / code adjustments for WASM & Windows.

[Barre]

I think this needs fixes / code adjustments for WASM & Windows.

My most recent changes should do the trick.

[crepererum]

Needs a rebase, otherwise looks good to me. @tustvold any objections?

[tustvold]

LGTM, just a few comments

• I think the namespace for the attributes should be user.object_store or possibly user.os
• I think we should probably feature flag this, just in case there are compatibility issues
• I do wonder what the performance implications of this are

[Barre]

LGTM, just a few comments

• I think the namespace for the attributes should be user.object_store or possibly user.os
• I think we should probably feature flag this, just in case there are compatibility issues
• I do wonder what the performance implications of this are

Should be ok now.

[crepererum]

Isn't this missing the xattr feature?

[Barre]

Ah, I thought you didn't want it by default.

[crepererum]

Yes, but you still need to define the feature. An optional dependency doesn't automatically define a feature, no?

[Barre]

$ cargo metadata --format-version 1 --no-deps | jq '.packages[0].features'

[...]
  "xattr": [
    "dep:xattr"
  ]

[crepererum]

Weird, but since it passes the test this is obviously working 🤷

[Barre]

(Ran cargo fmt on last force-push).

[crepererum]

I think we need to do something about Windows. So I think selecting this feature on Windows should probably lead to compile_error! and the feature shouldn't be part of default = [...] (since we don't support it on all platforms).

Or you find a way to set file attributes on Windows (see for example this discussion), but I think this might be a bit out of scope for this PR.