Use time.monotonic() instead of time.time() for elapsed time calculations

[veeceey]

Summary

• Replaces all 92 uses of time.time() with time.monotonic() across 18 source files for measuring elapsed time, timeouts, and latency
• Updates 7 test files to mock time.monotonic instead of time.time, ensuring all unit tests continue to pass
• Preserves time.time() in kafka/record/legacy_records.py where wall clock time is needed for Kafka message timestamps

Motivation

time.time() returns wall clock time which is adjustable -- it can jump forward or backward due to NTP synchronization, DST changes, or manual clock adjustments. Using it for elapsed time calculations can cause:

• Premature timeouts: if the clock jumps forward, operations may time out before their actual deadline
• Infinite hangs: if the clock jumps backward, timeout conditions may never be satisfied
• Incorrect latency metrics: recorded latencies can be negative or wildly inflated

time.monotonic() provides a clock that is guaranteed to never go backward, making it the correct choice for all interval/elapsed time measurements.

>>> import time
>>> time.get_clock_info('time')
namespace(adjustable=True, implementation='clock_gettime(CLOCK_REALTIME)', monotonic=False, ...)
>>> time.get_clock_info('monotonic')
namespace(adjustable=False, implementation='clock_gettime(CLOCK_MONOTONIC)', monotonic=True, ...)

Fixes #1546

Test plan

• All existing unit tests pass (1656 passed, 28 skipped -- the 1 pre-existing failure in test_crc32c[None] is unrelated)
• Updated test mocks in test_metrics.py, test_client_async.py, test_conn.py, test_coordinator.py, test_sender.py, test_fetcher.py, and testutil.py to patch time.monotonic instead of time.time
• Verified kafka/record/legacy_records.py still uses time.time() for Kafka message timestamps (wall clock time is correct there)

🤖 Generated with Claude Code