That's my answer; neat!
This is a contentious issue for musicians. You're right that beat 3 is a i
chord in second inversion; it has C–E♭–G
, obviously a C-minor chord, and that should be i
!
But let's imagine that the E♭–G–C
on beat 3 didn't exist, and we just had a half note of D–G–B
. We'd label that V
, right? Well, this notation just shows that the V
chord really "begins" on beat 3, we just have some accented non-chord tones up above. The C
is a type of 4–3 suspension, the G
already fits, and the E♭
is basically an accented passing tone. Especially when we have that G
in the bass, our ears are already conceptualizing that beat 3 as a dominant chord. (It also helps that this particular resolution is one of the most common things in common-practice harmony; we call it the "cadential six-four.")
So to show that this is just an ornamented V
chord, we go ahead and label it with a V
Roman numeral, even though technically that beat 3 has all the notes of a i
chord. Roman-numeral analysis tries to show how a chord functions, and since beat 3 is really the start of the dominant, we go ahead and use a V
Roman numeral.
The superscript numbers just show how the upper voices are moving. The 6–5
indicates that a sixth above the bass (E♭
) moves to a fifth above (D
); likewise, the 4–3
shows the movement C–B
.
Other methods of analysis would be:
I64–V
, as you said. This is very common, but it fails to show how beat 3 is already the dominant.
I64–V
, but with a bracket below, labelled V
, showing that both of those chords are really functioning as dominant. This, in my opinion, is the clearest approach.
PS - Something very similar happens in the next-to-last measure.