[qob] remove resource leak in readNoCompression

[danking]

CHANGELOG: In Azure Query-on-Batch, fix a resource leak that prevented running pipelines with >500 partitions and created flakiness with >250 partitions.

fixes: #13061

[danking]

I verified this scales all the way to 50,000 partitions but the batch-driver can't schedule these fast enough to make the test fast. They take less than a second but more than 300ms. We'd need like 64 8-core (512 cores) nodes to bring test time down to a reasonable amount. We should strive to get there but batch would need to schedule at 512 jobs per second for that to make sense. Hmm, something went wrong.

OK, we'll need to revisit 50k partition tables. But let's get this in, the current code is obviously wrong.

[daniel-goldstein]

@danking Gotta regenerate the dev pinned requirements.

[danking]

New ruff version found new issues which I fixed. Should pass now/