So now let's write a random vector $(x,y)$ in the basis $(1,0)$ and $(1,t-1)$
It equals $(x-\frac y{t-1})(1,0)+\frac y{t-1}(1,t-1)$
Probably I guess it's most useful to write $(0,1)$ in this basis
$(1,t-1)\frac1{t-1}-(1,0)\frac1{t-1}$
I guess the picture hear is that a shear is the limiting case of two eigenspaces getting closer and closer together
and therefore the algebraic multiplicity counts those two eigenspaces that "merged"
(or you could imagine it as two infinitesimally close eigenspaces)
(the geometric multiplicity counts the two "infinitesimally close" eigenspaces as the same, but the algebraic multiplicity is more discerning)