Earlier in the design, we accepted a race condition where the chunk cache could have stale content. The downside of this is the risk of a mismatched checksum. To avoid this taking down the entire program on a failed checksum from the chunk cache, the block manager should make one (and only one) attempt to:
1. Evict the chunk,
2. Retry a __wt_read.