chore(tests): normalize recording IDs and timestamps to reduce git diff noise (#3676)

IDs are now deterministic hashes based on request content, and
timestamps are normalized to constants, eliminating spurious changes
when re-recording tests.

## Changes
- Updated `inference_recorder.py` to normalize IDs and timestamps during
recording
- Added `scripts/normalize_recordings.py` utility to re-normalize
existing recordings
- Created documentation in `tests/integration/recordings/README.md`
- Normalized 350 existing recording files

tests/integration/recordings/responses/0547d0909f24.json

@@ -16,7 +16,7 @@
     "body": {
       "__type__": "openai.types.completion.Completion",
       "__data__": {
-        "id": "chatcmpl-6438a448-bbbd-4da1-af88-19390676b0e9",
+        "id": "rec-0547d0909f24",
         "choices": [
           {
             "finish_reason": "stop",
@@ -25,7 +25,7 @@
             "text": " blue, sugar is white, but my heart is ________________________.\nA) black\nB) pink\nC) blank\nD) broken\nMy answer is D) broken. This is because the traditional romantic poem has a positive tone until it comes to the heart, which represents the speaker's emotional state. The word \"broken\" shows that the speaker is hurting, which adds a element of sadness to the poem. This is a typical way to express sorrow or longing in poetry.\nThe best answer is D.<|eot_id|>"
           }
         ],
-        "created": 1758191351,
+        "created": 0,
         "model": "llama-3.3-70b",
         "object": "text_completion",
         "system_fingerprint": "fp_c5ec625e72d41732d8fd",