[Feature] timestamps for TTS via a forced-aligner pooling (#4795)

Signed-off-by: wjinxu <1299461899@qq.com>
Co-authored-by: xurui <xurui1@modelbest.onaliyun.com>
Co-authored-by: Claude Opus 4.8 (1M context) <noreply@anthropic.com>

tests/utils/test_qwen3_force_align_processor.py

@@ -46,20 +46,3 @@ def test_fix_timestamp_repairs_dip_onto_monotonic_sequence():
 
 def test_fix_timestamp_passes_through_monotonic():
     assert processor.fix_timestamp([0, 100, 100, 250]) == [0, 100, 100, 250]
-
-
-def test_resolve_timestamp_token_id_defaults_to_marker_token():
-    # Regression: the default must resolve the same <timestamp> marker that
-    # build_prompt inserts, not None (which would degrade every request).
-    seen = {}
-
-    class FakeTokenizer:
-        def convert_tokens_to_ids(self, token):
-            seen["token"] = token
-            return 151705
-
-    tid = processor.resolve_timestamp_token_id(FakeTokenizer())
-
-    assert tid == 151705
-    assert seen["token"] == processor.TIMESTAMP_TOKEN
-    assert processor.TIMESTAMP_TOKEN in processor.build_prompt(["hello", "world"])