CurledWake: use non-dimensional Ct in Gamma_0 (take 2)

[kirbyh]

The quoted text below is from @zanig in

---

The initial circulation Gamma_0 in CurledWakeWindfield.stamp_ic was using
the dimensionalized rotor.Ct (= sol.Ct * REWS^2) instead of the
non-dimensional rotor.extra.Ct (= sol.Ct).
After the fix, Gamma_0 scales linearly with U_REWS, matching both the
dimensional argument ([Gamma] = L^2/T) and the referenced Shapiro et al. (2018) form.

Also adds tests/test_uref_invariance.py: it runs a 2-turbine yawed farm with the
k-l turbulence model at U_0=1 and U_0=2 and asserts that REWS/U_0 at each
turbine is invariant under freestream rescaling, and that U_REWS <= U_0 for a Uniform Inflow.

To isolate the bug into a single factor, I ran a 4-turbine inline row (20° yaw, k-l model) at U_0=1 and U_0=8. Because $\Gamma_0 \propto U_{\rm REWS}^3$ instead of $U_{\rm REWS}^1$, each downstream turbine accumulates a stray U^2 factor and the ratio of $U_{\rm REWS}/U_0$ between the two runs diverges from 1.0, shown below.

Turbine	current (rotor.Ct)	fixed (rotor.extra.Ct)
1	1.0000	1.0000
2	4.0293	0.9995
3	2.0580	1.0001
4	2.3485	1.0000

Turbine 1 has no upstream wake so it's invariant either way; T2–T4 sit downstream and pick up the $U_{\rm REWS}^2$ factor. Based on this observation is also the added test.

---

I've confirmed that the fix to use rotor.extra.Ct is indeed correct, and also made a small normalization fix to compute_x0_with_TI, which was introduced after the original PR.

This pull request also surrounds the package-level import of .FlorisInterface in a try/except block because floris is an optional dependency which may not be installed by all users.