Relax WriteMultipart API to support aborting after completion

[fsdvh]

In the current implementation finish will consume our writer and in case of an error during the finish call we wound be able to abort our write.

I suggest changing the following:

pub async fn finish(mut self) -> Result<PutResult>

to something like this:

pub async fn finish(&mut self) -> Result<PutResult>

To be able to call the abort method after that

We potentially can safeguard ourselves from multiple finish calls by adding an is_finished flag to the writer

[tustvold]

I think it is generally assumed that if finish errors the upload will be aborted and eventually cleaned up - https://docs.rs/object_store/latest/object_store/aws/index.html#multipart-uploads.

I think we could probably document more clearly that abort is an optimisation. We could also make WriteMultipart abort on failure

Edit: I'm pushing back on this largely out of a desire to avoid breaking API changes, which are problematic

[fsdvh]

So I made a change to make a cleanup automatic process during complete, wdyt @tustvold?

pub async fn finish(mut self) -> Result<PutResult> {
        if !self.buffer.is_empty() {
            let part = std::mem::take(&mut self.buffer);
            self.put_part(part.into())
        }

        self.wait_for_capacity(0).await?;

        match self.upload.complete().await {
            Err(e) => {
                self.tasks.shutdown().await;
                self.upload.abort().await?;
                Err(e)
            }
            Ok(result) => Ok(result),
        }
    }

[alamb]

label_issue.py automatically added labels {'object-store'} from #5974