If we did not make the final reflection, point multiplication would not be associative, we would not have a group, and we thus could not define a scalar multiplication the way we do, with the property $n\cdot A+m\cdot A=(n+m)\cdot A$ .
If operation $+$ is a group law, the operation $\boxplus$ defined on that group by $A\boxplus B=-(A+B)$ (which is what removing the reflection does) is generally not associative, because $$(A\boxplus B)\boxplus C=-(-(A+B)+C)=A+B-C,$$ while $$A\boxplus(B\boxplus C)=-(A-(B+C))=-A+B+C.$$ The only groups where $\boxplus$ is associative are those with 1 or 2 elements.
Associativity of point addition on an elliptic curve in fact is a non-trivial and fragile property. Messing with how we do point addition in almost any way (changing sign as proposed, using a curve with a different equation like an astroid..) breaks that property.
Comments by entrop-x and Rosie F provide an intuitive explanation: if three points $A$, $B$, $C$ on the curve are such that "point addition of $A$ and $B$ without final reflection yields $C=A\boxplus B$", then that statement holds for all 6 permutations of $A$, $B$ and $C$, and it can be reworded as "$A$, $B$ and $C$ are collinear" without consideration of the order of the points. In a group with law $+$, the simplest relation between 3 variables with that invariance under order is: $A+B+C=0$, and then we have $A+B=-C=-(A\boxplus B)$ (where the $-$ sign designates taking the opposite for law $+$ ), hence the final reflection to change $\boxplus$ into a group law $+$.
That intuitive explanation can be extended to justify that full point multiplication is associative!