Implement custom libevent adapter to fix timeout issues with event prioritization

[plainbanana]

Summary

This PR addresses timeout-related issues in Redis::Cluster::Fast by implementing a custom libevent adapter with event
prioritization.

Key Changes

• Custom libevent adapter: Added src/adapters/libevent.h to handle timeout issues specific to our use case
• Non-blocking event loop: Made run_event_loop truly non-blocking with EVLOOP_NONBLOCK
• Enhanced error handling: Added proper error checks for event_base_priority_init
• Documentation improvements: Fixed typos and clarified method descriptions
• Code quality: Added explanatory comments and improved test clarity

Problem Background

When using hiredis-cluster's async context, we encountered situations where timeout events would fire even when Redis
responses were already available to read. This particularly affected scenarios where event loop execution was delayed due
to system load.

Approach Taken

After researching the issue and consulting with the hiredis community hiredis issue
#1286 , we learned that:

1. libevent's event processing order: The library processes timeouts before I/O events as part of its design
2. Upstream perspective: The hiredis maintainers appropriately noted that adding event prioritization to the
   general-purpose adapter could introduce starvation risks in multi-client scenarios
3. Valid concern: In shared event base environments, high-traffic clients could potentially starve failing clients
   from receiving necessary timeouts

Our Solution

Given Redis::Cluster::Fast's specific usage patterns and requirements, we decided to maintain our own libevent adapter:

Why this works for our use case

• Isolated event base: Redis::Cluster::Fast typically manages dedicated event bases
• Controlled environment: Our specific client patterns reduce starvation risks
• User benefit: Processing available responses before timeouts significantly improves user experience

Implementation Details

Our custom adapter (src/adapters/libevent.h) implements:

• Separate event management: Independent handling of I/O and timer events
• Event prioritization: I/O events (priority 0) before timer events (priority 1)
• Deterministic behavior: Ensures available responses are processed before timeout evaluation

  // Priority 0: I/O events (highest priority) - Process Redis responses immediately
  // Priority 1: Timer events (lower priority) - Handle timeouts after I/O processing
  #define EVENT_BASE_PRIORITY_NUMBER 2

Test Plan

• All existing tests pass
• Memory leak tests (xt/02_leak.t, xt/08_leak_srandom.t)
• Timeout handling tests (xt/04_timeout.t, xt/10_timeout_srandom.t)
• Valgrind tests (xt/05_valgrind.t, xt/09_valgrind_srandom.t)
• Fork safety tests (xt/03_fork.t, xt/11_fork_srandom.t)
• Pipeline functionality verification

This change improves reliability for Redis::Cluster::Fast users while respecting the design decisions of the upstream
hiredis project.