Summary

This PR refactors the Http class to remove its persistent HTTP::Client usage and instead delegate the creation of HTTP clients to a new factory interface (HttpClientFactory). By doing so, each request can optionally acquire a client from a custom provide.

What Changed

1. Removed the single @client field in Http:

• Previously, Http stored one HTTP::Client instance and reused it for all requests.
• Now, Http does not store any long-lived client. Instead, it calls a factory’s #acquire_client method on every request.

2. Introduced HttpClientFactory and DefaultHttpClientFactory:

• HttpClientFactory is an abstract class with three key methods:
  • #acquire_client(endpoint, signer) – returns a fully configured HTTP::Client.
  • #acquire_raw_client(endpoint, signer) – returns a plainHTTP::Client, must be implemented by developer.
  • #release(client) – allows returning or cleaning up the client. (Default implementation is no-op.)
• DefaultHttpClientFactory provides the current “create a new client each time” logic, so no immediate behavioral change occurs unless users swap in a different factory.

3. New request lifecycle in Http:

• Each HTTP request (e.g. delete, get, head, etc.) uses acquire_client to get a ready-to-use client.
• After the request finishes (or fails), Http calls release(client) to allow the factory to clean up or recycle resources.

How the New Mechanism Works

1. Http obtains an HTTP::Client on demand:

client = @factory.acquire_client(@endpoint, @signer)

2. Request Execution:

• The #exec method sends the HTTP request using the acquired client.
• Upon success or failure, we exit the loop and proceed to the ensure block.

3. Release:

• @factory.release(client) is called, allowing the factory to decide whether to close, reuse, or discard the client.

Closes #77