For the Bayesian linear regression, likelihood function is Gaussian, conjugate prior is Gaussian, we can get the posterior is also Gaussian.
However, for Bayesian logistic regression, likelihood is logistic sigmoid function, since evaluation of posterior distribution requires normalization of the product of prior and likelihood, it is intractable. Also, we can not just choose one theta, MAP(theta). Because in bayesian theory, theta is a random variable, which can have multiple value, we can not use one specific value. That is why we integrate theta.