TPU.scp_to() has hardcoded 300s timeout — no way to override for large transfers

Start in tpuz/tpu.py at scp_to around line 360, then inspect _gcloud around line 116 and the related scp_from, push, and pull entry points. Trace how each transfer invokes gcloud and verify that the chosen timeout behavior is consistently exposed through the public methods without changing unrelated command handling.

TPU.scp_to() in tpuz/tpu.py:360 calls self._gcloud(args, timeout=300) with a hardcoded 300 second timeout. There's no way to pass a longer timeout through the public API, so any transfer larger than roughly 100–200 MB at typical residential upload speeds hits subprocess.TimeoutExpired even though the underlying gcloud scp is making forward progress.

Repro

from tpuz import TPU
tpu = TPU("some-tpu", accelerator="v6e-8", zone="europe-west4-a", preemptible=True)
tpu.up_queued(timeout_hours=2)
# flaxchat repo is ~1 GB (mostly models/ and exports/). scp_to will time out at 300s:
tpu.scp_to("/Users/me/code/flaxchat/", "/home/me/workdir/flaxchat")

Traceback

File ".../tpuz/tpu.py", line 360, in scp_to
    self._gcloud(args, timeout=300)
File ".../tpuz/tpu.py", line 116, in _gcloud
    result = subprocess.run(cmd, capture_output=True, text=True, timeout=timeout)
...
subprocess.TimeoutExpired: Command '['gcloud', 'compute', 'tpus', 'tpu-vm', 'scp',
'/Users/me/code/flaxchat/', 'some-tpu:/home/me/workdir/flaxchat',
'--zone=europe-west4-a', '--worker=0', '--recurse']' timed out after 300 seconds

Fix

Add a timeout kwarg to scp_to/scp_from/push/pull:

def scp_to(self, local, remote, worker=0, timeout=300):
    ...
    self._gcloud(args, timeout=timeout)

Also consider defaulting to None (no timeout) for scp, since transfer time is inherently unpredictable and the operation is interruptible. Timeouts make sense for ssh where a hang indicates a dead command, not for bulk data movement.

Environment

• tpuz==0.1.11
• Python 3.13 on macOS (client), TPU v6e-8 europe-west4-a (remote)

Workaround

Use git clone on the TPU for code, and scp_to only for small untracked files:

tpu.ssh("git clone --depth 1 https://github.com/org/repo /home/me/workdir/repo", timeout=180)
tpu.scp_to("/local/untracked_script.py", "/home/me/workdir/repo/scripts/untracked_script.py")