Some session recordings terminate prematurely

**Description**
I find that some of my session recordings cut off prematurely. This applies to 4 of 30 sessions in the study with UUID 951f3e6a-437c-46e5-a0c8-755b0b386446 (Babies, Blobs & Barriers). This doesn't seem to be related to the recording having reached a maximum file size, as the videos that end prematurely have smaller file sizes than many other videos that didn't have this issue. I haven't received feedback that the experiment stopped prematurely, and I get trial information from lookit for trials outside the recording, so I suspect that the experiment continued while the recording stopped.

**How to reproduce**
I detected these cases by trying to line up automated gaze coding of the videos with the time stamps provided by the lookit json output. For these videos, some trial time stamps were outside the total duration of the video.

**Expected behavior**
I would expect the duration of the video to equal the event timestamps: t(stopSessionRecording) - t(startSessionRecording). This is the case for most of the videos but for 4 of them.

** Environment (OS, browser, branch, etc.):**
I don't know which browser the parents used - I use Chrome on macOS.

**Additional context**
One suggestion is to move to trial-level, rather than session-level video. I haven't tried this yet. Prefer to avoid since it makes file management associated with automated gaze coding harder. But open to this being the only solution. Happy to help investigating in any way.

Start by investigating how session-level recordings are started and stopped, then compare recording durations with the event timestamps for study 951f3e6a-437c-46e5-a0c8-755b0b386446. Done means recordings consistently span from startSessionRecording to stopSessionRecording without ending before the final trial event; the issue names no files or tests to run.

Written by the indexing model from the issue text.