storage: mvccstats inconsistencies during tpcc

[tbg]

Running ./cockroach debug check-store against nodes from #28918, I see mismatches like this:

Note in particular r137 below, whose stats are pure garbage.
At first glance, it looks like this might be spillover from the timeseries ranges (which have approximate stats that tend to go negative when GC happens) but we're supposedly tracking the ranges which hold estimates and the ranges below claim not to.

We should generally be able assert that there isn't a stats mismatch and do so in all roachtests at opportune times.

r127:/Table/54/1/9{27/3/-2045/0-64/2/-1444} [(n1,s1):1, (n3,s3):2, (n2,s2):3, next=4, gen=842355161184]: diff(actual, claimed): GCBytesAge: 9090058794 != 9427657411
LiveBytes: 48418643 != 48349513
LiveCount: 1117871 != 1116225
KeyBytes: 25796071 != 25777965
KeyCount: 1117871 != 1116225

r402:/Table/54/{1/964/2/-1444/0-2} [(n1,s1):1, (n3,s3):2, (n2,s2):3, next=4, gen=842353299136]: diff(actual, claimed): GCBytesAge: 8582510619 != 9048690353
LiveBytes: 47015452 != 46930069
LiveCount: 1085445 != 1083412
KeyBytes: 25046213 != 25023850
KeyCount: 1085445 != 1083412

r137:/Table/55/1/{0/0-1000} [(n1,s1):1, (n3,s3):2, (n2,s2):3, next=4, gen=842350802872]: diff(actual, claimed): GCBytesAge: 459022290695 != 468795486289
LiveBytes: 82579 != -846349
LiveCount: 1000 != -12823
KeyBytes: 4112650 != 4021050
KeyCount: 1000 != -12823

r203:/Table/56/1/{0/1/2101/0-138/8/2871} [(n1,s1):1, (n3,s3):2, (n2,s2):3, next=4, gen=842354018792]: diff(actual, claimed): GCBytesAge: 75133733043 != 75467250960
LiveBytes: 32065065 != 31993943
LiveCount: 1222845 != 1220121
KeyBytes: 29396707 != 29371893
KeyCount: 1327389 != 1324665

r5553:/Table/56/1/{138/8/2871-272/10/2564} [(n1,s1):1, (n3,s3):2, (n2,s2):3, next=4, gen=0]: diff(actual, claimed): GCBytesAge: 40028914124 != 40406514945
LiveBytes: 32306114 != 32213674
LiveCount: 1190224 != 1186820
KeyBytes: 28209986 != 28175414
KeyCount: 1245160 != 1241756

[tbg]

We've since fixed multiple inconsistencies that would result in such negative stats being generated, see for instance #36861.