fix: Use perf_counter_ns for request duration tracking (#1260)

### Description

Replace `datetime.now` by `time.perf_counter_ns` in time difference
calculations for request duration tracking to avoid possible issues
connected to measured time resolution.
This fixes the flakiness of `test_final_statistics` that was happening
on Windows and Python versions < 3.13.

### Issues

- Closes: #1256

---------

Co-authored-by: Vlada Dusek <[email protected]>

Author: Pijukatel

src/crawlee/statistics/_statistics.py

@@ -2,6 +2,7 @@
 from __future__ import annotations
 
 import math
+import time
 from datetime import datetime, timedelta, timezone
 from logging import Logger, getLogger
 from typing import TYPE_CHECKING, Generic, Literal
@@ -27,22 +28,22 @@ class RequestProcessingRecord:
     """Tracks information about the processing of a request."""
 
     def __init__(self) -> None:
-        self._last_run_at: datetime | None = None
+        self._last_run_at_ns: int | None = None
         self._runs = 0
         self.duration: timedelta | None = None
 
     def run(self) -> int:
         """Mark the job as started."""
-        self._last_run_at = datetime.now(timezone.utc)
+        self._last_run_at_ns = time.perf_counter_ns()
         self._runs += 1
         return self._runs
 
     def finish(self) -> timedelta:
         """Mark the job as finished."""
-        if self._last_run_at is None:
+        if self._last_run_at_ns is None:
             raise RuntimeError('Invalid state')
 
-        self.duration = datetime.now(timezone.utc) - self._last_run_at
+        self.duration = timedelta(microseconds=math.ceil((time.perf_counter_ns() - self._last_run_at_ns) / 1000))
         return self.duration
 
     @property

tests/unit/_statistics/test_request_processing_record.py

@@ -0,0 +1,17 @@
+from datetime import timedelta
+
+from crawlee.statistics._statistics import RequestProcessingRecord
+
+
+def test_tracking_time_resolution() -> None:
+    """Test that `RequestProcessingRecord` tracks time with sufficient resolution.
+
+    This is generally not an issue on Linux, but on Windows some packages in older Python versions might be using system
+    timers with not so granular resolution - some sources estimate 15ms. This test will start failing on Windows
+    if unsuitable source of time measurement is selected due to two successive time measurements possibly using same
+    timing sample."""
+    record = RequestProcessingRecord()
+    record.run()
+    record.finish()
+    assert record.duration
+    assert record.duration > timedelta(seconds=0)