The previous way of initializing the s3 client in each call was
adding a lot of overhead and would tank performance beyond about
20 simultaneous requests.
Since the backend access is through a single account, we can init
and store this client for use from each api call.
This builds on the previous work that sets up the body streaming
for the put object and put part requests. This adds the auth and
checksum readers to postpone the v4auth checks and the content
checksum until the end of the body stream.
This means that the backend with start reading the data from the
body stream before the request is fully validated and signatures
checked. So the backend must check the error returned from the
body reader for the final auth and content checks. The backend
is expected to discard the data upon error.
This should increase performance and reduce memory utilization
to no longer require caching the entire request body in memory
for put object and put part.
This backend redirects incoming requests to another s3 service.
This will use the incoming credentials to setup the client
requests to the external s3 service. So the IAM accounts (or
root account) must match what the external s3 service expects.
