Add exponential backoff jitter to gRPC channel readiness

Background

C++ agent gRPC reconnect behavior currently relies on fixed channel readiness waits. When collectors are down or restarting, many agents can retry on similar intervals and create a thundering herd risk.

The Java agent mitigates this with ExponentialBackoffReconnectJob using:

• initialInterval: 3000ms
• multiplier: 1.2
• randomizationFactor: 0.3
• maxInterval: 30000ms

Scope

• Add a reusable exponential backoff helper with jitter.
• Apply jittered backoff only in the shared gRPC channel readiness path.
• Use WaitForStateChange during the backoff window instead of sleeping, with 1s slices.
• Call readyChannel() before metadata unary RPCs.
• Reset backoff after the channel becomes ready.
• Add unit tests for delay growth, cap, reset, jitter bounds, and metadata channel-not-ready behavior.

Validation

• cmake --build --preset debug-cached
• ctest --preset debug-cached

Start at the shared gRPC channel readiness path and inspect readyChannel(), WaitForStateChange, and the metadata unary RPC entry points. Use the Java agent's ExponentialBackoffReconnectJob values as reference, then run the listed CMake build and ctest presets. Done means backoff growth, cap, reset, jitter bounds, and channel-not-ready metadata behavior are covered by unit tests.