[v2] Move Content-Length validation from urllib3 to botocore

[aemous]

Notes:

• In [v2] Update urllib3 to 2.6.3 #9971, we upgraded our urllib3 bundled dependency to the next major version. This introduced a regression where urllib3 performs Content-Length validation instead of botocore, causing a different error to be raised instead of botocore's IncompleteRead. As of urllib3 v2, Content-Length validation is enforced by default, and there is a config option to opt-out (Enable enforce_content_length by default urllib3/urllib3#2514). botocore's Content-Length validation was implemented in 2013 when urllib3 had no support for Content-Length validation, so botocore built its own Content-Length validation. This PR opts out of urllib3's Content-Length validation so this check is handled by botocore (as it was pre-urllib3-v2). This fixes a bug in S3 streaming to stdout when an IncompleteRead is encountered.

Description of changes:

• Pass enforce_content_length=False to urllib3, to opt-out of content-length validation. This brings Content-Length validation back to urllib3. This fixes a bug with encountering an IncompleteRead while streaming an S3 object to stdout.

Description of tests:

• Added a blackbox test locally to test the S3 streaming download scenario. Verified that the expected retry request is made by the AWS CLI. Without this change, the retry is not made, and the streaming download terminates and surfaces an incomplete read error. Note: this blackbox test is not included in this PR, but it will be added in a future PR as part of the same stack as [v2] Add blackbox tests #10591.
• Successfully ran all test suites and CI (see GitHub Actions).

By submitting this pull request, I confirm that you can use, modify, copy, and redistribute this contribution, under the terms of your choice.

[aemous]

Keeping this in draft mode until we finalize relevant internal discussions.