Panic: out of memory when running replay from HTTP API

When running a live batch replay from the HTTP API, kapacitord takes up all the memory and crashes. This happens with any script.
For instance:
TICKscript:
batch
|query('''SELECT LAST(temperature) as temperature FROM "db"."autogen".data''')
.period(5m)
.every(5m)

When running HTTP API (/kapacitor/v1/replays/batch", json={'past': '5m', 'task': 'task'}) it will crash with out of memory
When running from CLI everything is ok: kapacitor replay-live batch -task task -past 5m

Start by reproducing the batch replay through the HTTP API endpoint /kapacitor/v1/replays/batch with the provided task, past window, and TICKscript, then compare it with the CLI command kapacitor replay-live batch -task task -past 5m. Trace the two replay entry points and identify why the HTTP path consumes unbounded memory; done means the API replay completes without an out-of-memory crash like the CLI path.

Written by the indexing model from the issue text.