Keep SVC spatial layer count and scaling factors across partial TUs

Previously, GetSvcParams derived number_spatial_layers and scaling
factors solely from the current temporal unit's frame_settings. When a
temporal unit omitted the top spatial layer(s), number_spatial_layers
shrunk for that TU and grew again on the next full TU, causing libaom
to reset SVC state and force a keyframe.

Track the high-water mark of spatial layers and scaling factors across
Encode() calls since InitEncode() so partial TUs preserve the full SVC
configuration.

Bug: webrtc:496266459
Change-Id: I86982578de18c3915e01bb70e8753a2f57810ba6
Reviewed-on: https://webrtc-review.googlesource.com/c/src/+/507842
Commit-Queue: Erik Språng <sprang@webrtc.org>
Reviewed-by: Sergey Silkin <ssilkin@webrtc.org>
Cr-Commit-Position: refs/heads/main@{#48820}