I don't believe that Bishop explicitly assumed all functions are continuous. I think that "no discontinuous function can be proved to be total in Bishop's constructive mathematics" is actually a very good way of summarizing the situation.
For example, consider how we would state "All functions from $\mathbb{R}$ to $\mathbb{Z}$ are constant" in a Bishop-type framework. A function would be a procedure that takes a representation of a real number and produces an integer, which is extensional in the sense that any two representations of the same real must produce the same integer. Now, if we try to produce a discontinuous function in this sense, we will find that the definition for the function cannot be proved to be total. This is because, if the function was total, then we could prove that for each real there is a finite amount of information about the real which already determines the output (using some version of Weak König's lemma, say), and that would mean that the function would need to be continuous. The key point is that it is not a function that is proved to be total, but rather a definition of a function.
In their Varieties of Constructive Mathematics (1987), which is a good reference on the subject, Bridges and Richman work with an non-formal system BISH which is a subsystem both of classical mathematics and of other constructive frameworks. In particular, every proof in BISH is also a proof in classical mathematics, so BISH cannot prove every function $\mathbb{R}\to\mathbb{N}$ is constant, or anything like that. They explicitly describe BISH as "Bishop's mathematics", and because Bridges had already been a co-author for the later edition of Constructive Mathematics by Bishop and Bridges, I take this as a relatively authoritative opinion on Bishop's viewpoint.
On the subject of partial functions, the system RUSS of Bridges and Richman adds a classically false axiom:
CPF: There is an enumeration $\phi_0$, $\phi_1$, $\ldots$ of all partial functions from $\mathbb{N}$ to $\mathbb{N}$ with countable domains.
They show that BISH + CPF does prove every function from $\mathbb{R}$ to $\mathbb{N}$ is constant, and so BISH + CPF is not compatible with classical models of analysis.
Nowadays, we sometimes identify BISH with formal systems of constructive arithmetic in all finite types (e.g. variants of $\text{HA}^\omega$). These systems are also compatible with classical logic, in the sense that the standard model of arithmetic is also a model of these theories. So these formal representations of BISH also don't prove every function is continuous.