chore(deps): replace unmaintained backoff crate with backon

[pranc1ngpegasus]

What was changed

The SDK depended on the unmaintained backoff crate (see RUSTSEC-2025-0012 / RustSec advisory) for exponential retry delays in:

• temporalio-client — gRPC call retry (TonicErrorHandler in retry.rs)
• temporalio-sdk-core — poller autoscaling error backoff (poll_buffer.rs)

This PR replaces backoff with backon in both crates:

1. temporalio-client: RetryOptions builds a backon::ExponentialBuilder; TonicErrorHandler drives retry delays via Iterator::next() on backon::ExponentialBackoff. Jitter is enabled when randomization_factor > 0 (with_jitter()). Long-poll fatal-error grace still uses wall-clock elapsed time via retry_started_at.
2. temporalio-sdk-core: Poll scaler backoff uses shared ExponentialBuilder constants aligned with RetryOptions::task_poll_retry_policy() and throttle_retry_policy(); state resets with builder.build() after successful polls.

Why?

Fixes #1292. RUSTSEC-2025-0012 reports that backoff is unmaintained and recommends migrating to backon for exponential backoff/retry. Each crate now depends on backon directly so sdk-core does not pull retry utilities through client.

Checklist

1. Closes [Maintenance] Replace unmaintained backoff crate with backon (RUSTSEC-2025-0012) #1292

2. How was this tested:

   • cargo test -p temporalio-client retry::
   • cargo test -p temporalio-sdk-core autoscaler_applies_backoff

3. Any docs updates needed?
   N/A — internal dependency and retry implementation change only; public RetryOptions fields are unchanged.