fix(framework): Make retry waits cancellable

[tanertopal]

What changed

• add an explicit RetryInvoker.disable_retries() API
• allow retry implementations to interrupt an active wait
• re-check retry limits after an interrupted wait
• protect mutable retry state with reentrant synchronization

Why

Lowering max_tries does not wake an invocation that is already sleeping in backoff. Shutdown needs an explicit operation that prevents new retries and interrupts current waits.

Stack

Umbrella/reference: #7895 — Prevent task executor shutdown deadlock

• Previous: #7914 — Make exit handling bounded and reentrant
• This PR: fix(framework): Make retry waits cancellable #7915 — Make retry waits cancellable
• Next: #7916 — Cancel gRPC retries during shutdown

Merge the stack in order. After the previous PR merges, retarget this PR to main.

Validation

• pytest py/flwr/supercore/retry/retry_invoker_test.py
• Ruff and Black on the changed retry modules
• git diff --check