feat: add ModelThrottleError for rate limiting

[lseguin1337]

Motivation

When model providers throttle requests due to rate limiting, developers need a consistent way to detect and handle these scenarios. This enables implementing retry strategies, backoff mechanisms, or graceful degradation in applications.

Changes

Introduces ModelThrottleError, a new error type thrown when rate limiting is detected:

Before:

// Bedrock: Error was thrown as raw event data
// OpenAI: Error was propagated as-is, requiring inspection of error properties

After:

import { Agent, ModelThrottleError } from '@strands-agents/sdk'

try {
  for await (const event of agent.run('Generate a long story')) {
    // handle events
  }
} catch (error) {
  if (error instanceof ModelThrottleError) {
    // Implement retry logic with backoff
    console.log('Rate limited, retrying after delay...')
  }
}

Provider-specific behavior

• BedrockModel: Detects throttlingException events during streaming
• OpenAIModel: Detects HTTP 429 status, rate_limit_exceeded error codes, and common rate limit message patterns

Closes #424

[pgrayy]

Can you elaborate on the purpose of { cause: err } and why you pass it in here but not for BedrockModel?

[lseguin1337]

the goal was the forward the original error, but maybe it's not necessary. Let me know I can remove it