Scan sequences of unknown fields and add as a batch whenever an unknown field is encountered in the fast decoder.

The most likely thing to come after an unknown field is another unknown field, either because of treeshaking (where most fields are unused) or schema evolution (new fields added after known fields) or extensions (large tag numbers).

This basically eliminates minitable searches and unknown field additions as a bottleneck.

PiperOrigin-RevId: 919139292