Inconsistent results for a definite integral

Consider the integral

\frac{\int_{b-i \infty }^{b+i \infty } \frac{\exp (a z)}{\sin (\pi z)} \, dz}{2 \pi i}

where 0 < b <1 and |Im a| < 1. Below is Mathematica code which evaluates this integral 1) directtly and 2) evaluating the indefinite integral between the two limits.
The Mathematica command:

value1=Assuming[0 < b < 1 && Abs[Im[a]] < Pi,1/(2 Pi I) Integrate[Exp[a z]/Sin[Pi z],{z,b- I Infinity,b+I infinity}]] 

produces the result

-\frac{1}{\pi \left(e^a+1\right)}

If we calculate the indefinite integral

indef=1/(2 Pi I) Integrate[Exp[a z]/Sin[Pi z],z] 

we obtain

\frac{i e^{(a+i \pi ) z} \, _2F_1\left(1,\frac{\pi -i a}{2 \pi };\frac{3}{2}-\frac{i a}{2 \pi };e^{2 i \pi z}\right)}{\pi (\pi -i a)}

Next, we set

temp = FullSimplify[(indef /. {z -> b + I R}) - (indef /. {z -> 
b - I R})] 

obtaining

-\frac{e^{(a+i \pi ) (b+i R)} \, _2F_1\left(1,\frac{\pi -i a}{2 \pi };\frac{3}{2}-\frac{i a}{2 \pi };e^{2 i \pi (b+i R)}\right)-e^{(a+i \pi ) (b-i R)} \, _2F_1\left(1,\frac{\pi -i a}{2 \pi };\frac{3}{2}-\frac{i a}{2 \pi };e^{2 \pi (i b+R)}\right)}{\pi (a+i \pi )}
value2 = Assuming[0 < b < 1 && Abs[Im[a]] < Pi && R > 0, 
  Limit[temp, R -> Infinity]]] 

gives

\frac{e^{a/2} \Gamma \left(\frac{3}{2}-\frac{i a}{2 \pi }\right) \Gamma \left(\frac{i a+\pi }{2 \pi }\right)}{\pi (\pi -i a)}
value2 = Factor[FullSimplify[value2]] 

gives

\frac{e^a}{\pi \left(e^a+1\right)}
{value1, value2}
\left\{-\frac{1}{\pi \left(e^a+1\right)},\frac{e ^a}{\pi \left(e^a+1\right)}\right\}

It can be shown by contour integration that value2 is the correct value.

Even in the simplest possible case, when a is 0, are there not singularities along the integration path, at each integral point on the imaginary axis? So in that case, what do you expect Mathematica to do with the first method? (Then there’s the separate issue of implicit limits, in two directions.)

It seems to me that even the second method is problematic, as the theorem relating a definite integral to antiderivative makes certain regularity assumptions along (and near) the path of integration. So for the second method, you’d have to take little semicircular detours around each of those singularities.

The integrand is an entire function except at the points where sin(Pi z) =0. i.e. z=0, ± 1, ± 2,…
There are no singularities on the path of integration, since along this path Re(z)=b and 0 < b <1.

Oops, sorry, I misread the argument of the integrand’s denominator.