Fix compressed-tensors loading for KV-cache-only quantized models (#47904)

The early return `if not remaining_groups: return` skipped
`apply_quantization_config` entirely when all config groups were
handled by the FP8 kernel path (or when there were no weight
quantization groups at all). This broke KV-cache-only quantized
models because `apply_quantization_config` is also responsible
for setting up KV-cache quantization observers and scales.

Co-authored-by: Claude Opus 4.6 <noreply@anthropic.com>