fix: Default AsyncPredictor upload prefix to endpoint name

[lucasjia-aws]

Issue

Fixes #3210
Fixes #4774

V3 counterpart: #6336

Problem

sagemaker.predictor_async.AsyncPredictor(predictor) defaults name to None, but calling predict(data=...) or predict_async(data=...) without an input_path fails with TypeError: 'NoneType' object is not subscriptable. This was reported on 2.92.1 (#3210) and again on 2.224.2 (#4774).

Root cause

AsyncPredictor._upload_data_to_s3 builds the S3 key with name_from_base(self.name, short=True), and name_from_base slices its base argument without a None guard. The existing unit test test_async_predict_call_with_data only passed because it set predictor_async.name manually before calling predict_async.

Fix

• In _upload_data_to_s3, use self.name or self.endpoint_name as the base for the S3 key prefix. An explicitly provided name is still used unchanged, and self.name is not modified.
• Documented the name argument in the AsyncPredictor.__init__ docstring.

Testing

• Added test_async_predict_call_with_data_and_no_name and test_async_predict_call_with_data_and_name_uses_name in tests/unit/test_predictor_async.py, using a real Predictor with a mocked session.
• The no-name regression test fails on the current master-v2 and passes with this change; tests/unit/test_predictor_async.py passes 22/22.
• flake8 passes on the changed files.

[mohamedzeidan2021]

Approving. Reviewed against the v2 tree on its own merits rather than as a parity check. Reproduced the issue's own snippet with a real Predictor: TypeError: 'NoneType' object is not subscriptable at utils.py:131 on master, correct S3 key and Accept after. 1 new test fails on reverted source; 22 passed, plus 62 across predictor/async_inference and 189 across the model tests.

Model.deploy passes a non-None name, so that path is untouched, and .name is read nowhere else in v2. The documented layout in doc/overview.rst is unchanged. The pre-existing integ test test_async_walkthrough passed against real AWS in this PR's CI, which is good no-regression evidence for the name-provided path (though nothing exercises name=None end to end).

Two non-blocking notes:

• self.name or self.endpoint_name treats name="" as unset; if self.name is None would be exact.
• tests/unit/test_predictor_async.py sets ENDPOINT and BUCKET_NAME both to "mxnet_endpoint", so the startswith("async-endpoint-inputs/{ENDPOINT}-") assertion can't distinguish "endpoint name was used" from "bucket name was used". It does still fail without the fix, so it earns its keep — it's just weaker than it reads. The v3 test gets this right with distinct values.

One sentence in doc/overview.rst noting the new default sub-prefix would close the loop. Note this overlaps files with #6334.