feat: implement retry mechanism for log processing

[kacpersaw]

Closes #119

This PR adds error handling and retry functionality to the log streaming by adding exponential backoff retry logic for failed log operations.

Changes

• Added retry mechanism with exponential backoff - automatically retries failed log operations, starting at 1 second delay and doubling up to 30 seconds max
• Retry logic for failed API operations - triggers retries for PostLogSource and Agent dRPC client failures to handle connection issues
• Per-agent retry state tracking - built retry state struct for each agent with individual timer and delay management
• Automatic retry cleanup - clears retry state on successful operations and proper cleanup on agent deletion
• Tests

With this PR, users will see more reliable logs from pod startup in their Coder workspace, including pod events before container startup that were previously not visible.

[deansheather]

What happens when the log never succeeds? Maybe because an agent token is no longer valid. Judging by the code, logstream will attempt to send it to coder Forever

[deansheather]

This will be kept in memory forever now right?

[kacpersaw]

It will be kept in memory until the pod is deleted.

https://github.com/coder/coder-logstream-kube/blob/kacpersaw/feat-log-queue/logger.go#L562