[diff] L2 method for Caputo derivatives is first order

[alexfikl]

The L2C method seems to be working correctly. Likely just an index issue or something? Needs a bit of debugging.

[alexfikl]

The L2 method applied to the Caputo derivative with $\alpha = 1.5$ looks like

h	Error	EOC
3.333e-02	6.659753e-02	---
1.613e-02	3.341454e-02	0.950
7.937e-03	1.690615e-02	0.961
3.937e-03	8.552690e-03	0.972
1.961e-03	4.318041e-03	0.980
9.785e-04	2.175312e-03	0.986
6.519e-04	1.455290e-03	0.990
4.888e-04	1.093799e-03	0.991
Overall		0.975

The L2 method applied to the Riemann-Liouville derivative with $\alpha = 1.5$ (through D_{RL} = D_C - F(a)) gives

h	Error	EOC
3.333e-02	3.686930e-02	---
1.613e-02	9.678625e-03	1.842
7.937e-03	2.593723e-03	1.857
3.937e-03	6.847805e-04	1.900
1.961e-03	1.774681e-04	1.937
9.785e-04	4.536133e-05	1.962
6.519e-04	2.033538e-05	1.976
4.888e-04	1.149314e-05	1.981
Overall		1.916

For the Caputo derivative, it's first-order regardless of $\alpha$, but for the RL derivative it seems to behave roughly like $\alpha + 0.5$ instead of the expected $3 - \alpha$. Note quite sure what to make of that.