Proposed mid-cycle update of typedefs.list

I happened to notice that the buildfarm's current typedefs list
adds quite a few names that were not previously being captured,
for example

@@ -48,10 +48,15 @@ AggPath
AggSplit
AggState
AggStatePerAgg
+AggStatePerAggData
AggStatePerGroup
+AggStatePerGroupData
AggStatePerHash
+AggStatePerHashData
AggStatePerPhase
+AggStatePerPhaseData
AggStatePerTrans
+AggStatePerTransData
AggStrategy
AggTransInfo
Aggref

This is great, because it means that the declarations of these
structs need not look funny anymore. But I am not quite sure why
this happened. It's not a BF tooling change as I first thought,
because multiple animals are reporting these names and the
same animals are not capturing these names on the back branches.
The best theory I can come up with is that 1b105f947 et al
used these names in palloc0_array and similar calls, and that
somehow looks like a capturable typedef reference ... but how?

Anyway, I'll gladly take this outcome. I propose applying the
attached to sync the in-tree typedefs list with what the
buildfarm is currently reporting.

One change I did not apply is that the buildfarm's list omits pgoff_t,
although we certainly still use that. This is evidently because
pgoff_t is defined as a macro not a typedef name. I guess we've been
manually preserving that name in the list, but it seems like we should
change "#define pgoff_t off_t" to "typedef off_t pgoff_t;" to avoid
that manual hack. I've not done that here, though.

regards, tom lane