It's easier with Lagrange I, i.e., you don't introduce the angle as a generalized parameter but work with (plane) Cartesian coordinates and imposing the constraint with a Lagrange parameter. The corresponding Lagrangian reads (setting ##\vec{x}=(x,y)##)
$$L=\frac{m}{2} \dot{\vec{x}}^2 -mg y +\frac{\lambda}{2} (\vec{x}^2-L^2).$$
The Euler-Lagrange equations read
$$m \ddot{x} = \lambda x, \quad m \ddot{y}=-mg + \lambda y.$$
From this it follows
$$\vec{T}=m \ddot{x}+m g \vec{e}_y = \lambda \vec{x}.$$
To get ##\lambda## note that
$$m x \ddot{x} + m (\ddot{y}+g) y = \lambda(x^2+y^2)=\lambda L^2 \; \Rightarrow \; \lambda = \frac{m(\vec{x} \cdot \ddot{\vec{x}}+g y)}{L^2}.$$
Now you introduce ##\theta## via ##\vec{x}=L(\sin \theta,-\cos \theta)##. Plugging this into the formula for ##\lambda## you find
$$\lambda=-\frac{m}{L} (L \dot{\theta}^2 + g \cos \theta).$$
Thus
$$\vec{T}=-m(L \dot{\theta}^2 + g \cos \theta) \begin{pmatrix}\sin \theta \\ -\cos \theta \end{pmatrix}$$
and
$$|\vec{T}|=m(L \dot{\theta}^2 + g \cos \theta).$$