Add managed SageMaker job connectors

[rohitrsh]

Summary

Expand flytekitplugins-awssagemaker beyond deployment and endpoint management with first-class async connectors for SageMaker managed jobs:

• Training
• Processing
• Batch Transform
• Hyperparameter Tuning
• Inference Recommender

Tracks flyteorg/flyte#7767.

Design

Each task follows Flyte's standard async connector lifecycle:

• create() submits the corresponding SageMaker job
• get() polls the describe API and maps service states to Flyte phases
• delete() stops the job and tolerates already-terminal or missing resources

The tasks reuse the plugin's existing boto3 templating for {inputs.X}, {images.X}, and {idempotence_token}. Successful describe responses are projected into stable result dictionaries so workflows can pass model artifacts, processing destinations, tuning results, recommendations, and transform output locations downstream without re-querying SageMaker.

The shared boto3 mixin now normalizes whole-number floats before request hashing/calls and preserves semantically significant list order. Duplicate-job recovery is only enabled when Flyte generated an idempotence token; quota and authorization failures remain visible to users.

No FlytePropeller changes are required. Each connector self-registers through the existing flytekit.plugins entry-point mechanism.

Package layout

The existing plugin gains five public subpackages:

• flytekitplugins.awssagemaker_training
• flytekitplugins.awssagemaker_processing
• flytekitplugins.awssagemaker_batch_transform
• flytekitplugins.awssagemaker_hyperparameter_tuning
• flytekitplugins.awssagemaker_inference_recommender

Each package contains its task surface, async connector, lifecycle metadata, and synchronous stop/describe helpers. The README documents configuration and composition patterns.

Test plan

• 93 SageMaker plugin unit tests pass locally
• Connector create/get/delete paths, terminal-state mapping, projected outputs, idempotent retry, quota propagation, and cleanup error handling are covered
• Package imports and connector registration verified
• Repository pre-commit hooks pass: Ruff, Ruff format, codespell, pydoclint, whitespace and debug-statement checks
• Managed job lifecycle exercised against AWS SageMaker for training, processing, tuning, recommendation, and batch-transform workflows
• Upstream CI

Follow-up

Flyte-native Pythonic execution for Training and Processing is intentionally excluded. That follow-up will run a regular @task function body inside the SageMaker container via ContainerEntrypoint and handle Flyte inputs.pb / outputs.pb / error.pb. Keeping it separate allows the conventional boto3/config connector surface to be reviewed independently.

/cc @samhita-alla @machichima @kumare3 @pingsutw

[codecov]

Codecov Report

✅ All modified and coverable lines are covered by tests.
✅ Project coverage is 83.58%. Comparing base (ca59398) to head (ce5cbe9).

Additional details and impacted files

@@            Coverage Diff             @@
##           master    #3456      +/-   ##
==========================================
+ Coverage   81.95%   83.58%   +1.63%     
==========================================
  Files         217        3     -214     
  Lines       22943      195   -22748     
  Branches     3036        0    -3036     
==========================================
- Hits        18804      163   -18641     
+ Misses       3248       32    -3216     
+ Partials      891        0     -891     

☔ View full report in Codecov by Harness.
📢 Have feedback on the report? Share it here.

🚀 New features to boost your workflow:

• ❄️ Test Analytics: Detect flaky tests, report on failures, and find test suite problems.

[rohitrsh]

CI is green for the SageMaker plugin and all other jobs. The only failure is the unrelated flytekit-greatexpectations matrix job because its external fixture now returns 404: https://raw.githubusercontent.com/superconductive/ge_tutorials/main/data/yellow_tripdata_sample_2019-01.csv. I do not have repository permission to rerun the failed job; please rerun that job when convenient.