TRT-2288: Add configurable retry strategy for test failures (#30223)

* Add configurable retry policy system with multi-retry support

Introduces --retry-strategy flag with options "none", "once" *(default,
current strategy), and "aggressive". Aggressive retry mode runs all
failed tests up to 10 attempts each to confirm failures are real. Tests
with <4 failures across attempts are allowed to pass as flaky; >=4
failures create rollup failure test case with concatenated outputs from
all attempts.

* Adjust thresholds and add some unit tests

* Fix bug in once strategy

* Address review feedback

* Set default to aggressive for testing temporarily

* Add autodl file for tracking retries

* Revert "Set default to aggressive for testing temporarily"

This reverts commit db60b72.

Author: stbenjam