feat(pipeline): ✨ Add capability detection logic to validate video/system capabilities before processing

Co-Authored-By: Lilith Autocommit <noreply@atlilith.com>

services/imajin-video/service/src/pipeline/capability.py

@@ -0,0 +1,31 @@
+"""Server-capability sizing for concurrent video work.
+
+The throttle on how many videos imajin-video decodes at once lives HERE, on the
+server — never on the client that requests classification. A backfill consumer may
+fire dozens of jobs; this is what keeps apricot from being swamped.
+
+Pure + unit-tested: the derivation takes the host's available RAM and a per-video
+budget and returns a bounded concurrency, so it tests without a host or a GPU.
+"""
+from __future__ import annotations
+
+
+def derive_video_concurrency(
+    available_bytes: int,
+    per_video_budget_bytes: int,
+    ceiling: int,
+    *,
+    floor: int = 1,
+) -> int:
+    """How many videos to decode concurrently, bounded by host RAM.
+
+    Returns ``ceiling`` clamped down to however many ``per_video_budget_bytes``
+    chunks fit in ``available_bytes`` — never below ``floor``. A non-positive
+    budget disables the RAM clamp (returns ``max(floor, ceiling)``).
+    """
+    if ceiling < floor:
+        ceiling = floor
+    if per_video_budget_bytes <= 0:
+        return ceiling
+    by_ram = available_bytes // per_video_budget_bytes
+    return max(floor, min(ceiling, int(by_ram)))