Add an opt-in settings.watching.batch_size to fetch the initial listing (the list preceding every watch) in limit/continue pages instead of one large request, capping the cold-start peak-memory spike for high-cardinality cluster-wide watches (e.g. Pods/Jobs).

• list_objs() paginates via limit/continue when batch_size is set, aggregating items across pages.
• Keeps the snapshot resourceVersion (consistent across chunks) to start the watch from the fully-listed state.
• Fully backward compatible: batch_size defaults to None → single unpaginated request, unchanged URL/params.
• tests/k8s/test_list_objs.py (kmock-based) covers the default single request, multi-page aggregation + query params, empty-continue stop, and kind/apiVersion defaulting across pages.

Why

list_objs() issues one unpaginated GET and parses the entire collection into memory at once before the watch begins. For operators/observers watching high-cardinality built-in resources cluster-wide, the cold-start list materializes every object simultaneously — a peak-memory spike scaling with total object count.

We hit this with a Prefect Kubernetes worker, which embeds a kopf observer watching all Pods/Jobs cluster-wide (prefect_kubernetes/observer.py): after a restart, the cold-start list of thousands of objects pushed the container past its memory limit and it crash-looped (OOMKilled). The Kubernetes API already supports chunked retrieval for exactly this (Retrieving large results in chunks), and client-go's reflector paginates its list by default — kopf just wasn't using it.

Example usage

@kopf.on.startup()
def configure(settings: kopf.OperatorSettings, **_):
    settings.watching.batch_size = 500   # page size; None (default) = single request