feat(sdk): align create_linear_retry_strategy with create_retry_strategy (parity with JS #652)

[SAY-5]

Closes #483.

What

create_linear_retry_strategy now takes a LinearRetryStrategyConfig (Duration-based) and gains max_delay capping, jitter (default FULL), and retryable_errors / retryable_error_types filtering, matching create_retry_strategy. The shared jitter, error-resolution/matching, and final delay clamp logic is extracted into module-level helpers used by both helpers so they cannot drift.

Why

create_retry_strategy was already config-shaped with those capabilities; the linear helper lagged behind and forced callers to hand-roll a strategy for jitter, a delay cap, or error filtering. This mirrors the JS change in aws/aws-durable-execution-sdk-js#652.

Notes

This is a breaking change, as called out in the issue: the signature moves from positional (max_attempts, initial_delay, increment) to a config object, and a bare create_linear_retry_strategy() now applies FULL jitter. RetryPresets.linear() is pinned to jitter=NONE so its documented deterministic 1s-5s delays are preserved.

Testing

Updated the existing linear tests to the config API and added coverage for the new max_delay cap, jitter application, and message/type error filtering. hatch run test:cov passes locally for tests/retries_test.py.

[yaythomas]

please exclude from commit :-)

[yaythomas]

this could move to JitterStrategy and then it's nicely encapsulated in the Jitter object...

   # on JitterStrategy
   def finalize_delay(self, base_delay: float) -> int:
       return max(1, math.ceil(self.apply_jitter(base_delay)))

[SAY-5]

Dropped uv.lock from the branch and moved the finalize step onto the strategy as JitterStrategy.finalize_delay, so both creators now just call config.jitter_strategy.finalize_delay(base_delay). Tests still green.

[yaythomas]

great contribution, thank you so much @SAY-5! welcome to the repo 😁