refactor: Polish Writer API by merging append and write together

[Xuanwo]

This PR introduces the following changes:

• Remove append from OpWrite: this arg is confusing.
• Add content_length in OpWrite so that services can handle sized and unsized input by themselves

After all these changes, users only need to use write instead of append.

---

This change will make our writer's behavior more easy to understand.

[suyanhanx]

Seems like you're using write instead of append. Please append this change to desc.

[Xuanwo]

Seems like you're using write instead of append. Please append this change to desc.

DONE.

[suyanhanx]

LGTM on ci green.

[suyanhanx]

If users want to overwrite multiple times and then rollback, they will no longer be able to use the writer to do so.

WDYT?

[Xuanwo]

If users want to overwrite multiple times and then rollback, they will no longer be able to use the writer to do so.

I didn't understand these use cases. Could you provide more details?

[suyanhanx]

I didn't understand these use cases. Could you provide more details?

I think this is a problem of mixing write and append and not being able to distinguish between the two operations.

[Xuanwo]

No, we are confused many things together here.

After this PRs change, there is no append concept, we have two different write uses cases: Sized and Unsized.

If users specified the size for write operation and provide all data with it, we can write with oneshot. Otherwise, services need to handle it in different ways: append, multipart upload, range write, ...

op.write() is exactly this case: we know the size and we have all data.
op.writer() is the other case: we don't know the size in advance, and services need to handle it by different ways.

In general, we will hide all the impelment details inside services. The only thing for users to use is write, and the only choice for users is specify the size or not.

[suyanhanx]

No, we are confused many things together here.

After this PRs change, there is no append concept, we have two different write uses cases: Sized and Unsized.

If users specified the size for write operation and provide all data with it, we can write with oneshot. Otherwise, services need to handle it in different ways: append, multipart upload, range write, ...

op.write() is exactly this case: we know the size and we have all data. op.writer() is the other case: we don't know the size in advance, and services need to handle it by different ways.

In general, we will hide all the impelment details inside services. The only thing for users to use is write, and the only choice for users is specify the size or not.

Great explanation! I think this needs to be in the documentation.