test_in_tail: fix flaky throttling test

[Watson1978]

Which issue(s) this PR fixes:
Fixes #

What this PR does / why we need it:
The test "lines collected with throttling" was intermittently failing, especially on macOS CI runners.

4) Failure: test: lines collected with throttling(TailInputTest::throttling logs at in_tail level):
  elapsed_seconds 1.81999524999992 is out of allowed range:
    lower: 1.88 [sec]
    upper: 3.3200000000000003 [sec].
  <true> expected but was
  <false>
/Users/runner/work/fluentd/fluentd/test/plugin/test_in_tail.rb:2793:in 'block (3 levels) in <class:TailInputTest>'
     2790:           sleep(sleep_interval) while d.emit_count <= emit_count
     2791:           elapsed_seconds = Fluent::Clock.now - start_time
     2792:           if emit_count > 0
  => 2793:             assert_true(elapsed_seconds > lower_interval && elapsed_seconds < upper_interval,
     2794:                         "elapsed_seconds #{elapsed_seconds} is out of allowed range:\n" +
     2795:                         "  lower: #{lower_interval} [sec]\n" +
     2796:                         "  upper: #{upper_interval} [sec]")

The failure occurred because the test was measuring the test thread's observation delay rather than the actual interval between emit events.

Previously, start_time was recorded at the beginning of the whileloop. If the OS thread scheduling was delayed, the time spent waiting was excluded from the current cycle and bled into the next, causing elapsed_seconds to falsely fall below the lower_interval threshold.

By moving start_time = Fluent::Clock.now outside the loop and updating it immediately after confirming an emit, we accurately measure the duration between consecutive emits. This makes the test resilient to CI scheduling jitter.

Docs Changes:
N/A

Release Note:
N/A

[kenhys]

LGTM.