it will probably take a few tries until i get to the bottom of this, but so far this much is certain:
amsmath was derived from
amstex, by michael spivak.
\, was already in the
\aligned definition, and the adaptation to
amsmath didn't remove it.
i'm trying to unearth the (somewhat sparse)
amstex documentation to determine if it's explained anywhere. (this is akin to an archaeological dig.)
\null\, is endemic throughout definitions for anything in
that is matrix-like. from the (internal, undistributed) documentation:
\aligned@ is similar to
Before retiring from the AMS, I obtained permission to post the original
amstex documentation on CTAN. This consists of two files, both located in https://ctan.org/tex-archive/macros/amstex/doc :
later on, some particular features are addressed:
\matrix is very much like
\multilimits@, used before. However, now
there are two
\bgroups, corresponding to the
\endmatrix contributes two
There are two other new features. First, we put
\, around the
\vcenter; this looks better, especially with delimiters. The second
feature is the
\null before the first
\, This is because of the rule
regarding numbered formulas on TB, p. 189---if a very wide
used in a formula with a
\tag, we don't want TeX to assume that we put
glue before the
\matrix in order to control its positioning with
respect to that tag!
so one must reach the conclusion that the
\, is there on the assumption
that aligned structures will often(?)/usually(?) be used within delimiters,
as is usual with matrices.
it would have been reasonable to question this assumption during the
amsmath, but i can't find any evidence that this was done.
i'll put the matter onto the "research" list for the next overhaul of
amsmath, but (as usual) can't make any promises.