Panic when not able to delete compaction waste

[wi11dey]

• A failed rollback can leave behind a compaction product with data; if we continue to compact in this case, a tombstone can come in for some data in the failed product, expire, and leave no tombstone and no data, except in the product. The data in the failed product will be resurrected in the restart.
• A similar situation can occur if we fail to write to the write-ahead log and leave the ancestors after the view has shifted to the finalized products. These ancestors will not be wiped on the next restart because they were not written to the write-ahead log.

In both cases, we can prevent resurrection by refusing to compact after this case has been encountered, but this will cause the node to fall behind on compactions which will take longer to resolve than a bounce. We haven't observed either of these cases directly, which indicates that this failure mode is infrequent enough that a forced restart is an acceptable alternative to stopping compactions that doesn't risk extended perf degradation.