feat: add S3Stage downloader

[joshieDo]

• adds S3Stage skeleton logic. It's incomplete and not enabled. TODOs can be disregarded for now.
• adds S3Stage downloader. Downloads a file in parallel allowing resumes after shutdowns.

/// Downloads file from url to data file path.
///
/// If a `file_hash` is passed, it will verify it at the end.
///
/// ## Details
///
/// 1) A [`Metadata`] file is created or opened in `{target_dir}/download/{filename}.metadata`. It
///    tracks the download progress including total file size, downloaded bytes, chunk sizes, and
///    ranges that still need downloading. Allows for resumability.
/// 2) The target file is preallocated with the total size of the file in
///    `{target_dir}/download/{filename}`.
/// 3) Multiple `workers` are spawned for downloading of specific chunks of the file.
/// 4) `Orchestrator` manages workers, distributes chunk ranges, and ensures the download progresses
///    efficiently by dynamically assigning tasks to workers as they become available.
/// 5) Once the file is downloaded:
///     * If `file_hash` is `Some`, verifies its blake3 hash.
///     * Deletes the metadata file
///     * Moves downloaded file to target directory.

[mattsse]

cool,

I mostly have some questions re async or rather how we integrate the downloader in the s3stage

[mattsse]

this should probably emit some traces?

[mattsse]

have a few more questions about workers <> orchestrators

[mattsse]

do we need any additional header setup?

[joshieDo]

no, that's it. just need to know the total size of the file

[mattsse]

this will always be called when we call the poll function? this seems wrong, because we don't track whether we've downloaded everything

[joshieDo]

no, just when self.fetch_rx is None, meaning that there's no task in the background running. I should make it more explicit.

now that i look at it, there's an issue inside spawn_fetch that will always cause it to spawn, when it should only when theres files to be downloaded (maybe that's what you mean?)

[mattsse]

is this more performant than using a single downloader? I'd assume downloading is bound by bandwidth so unclear if using multiple downloaders improves this?

[joshieDo]

Although in theory it's bound by bandwidth, many times servers will throttle single connections. Using multiple connections helps

[mattsse]

this is a bit confusing to me,
why do we need multiple workers
doesn't this download just one file