http://en.wikipedia.org/wiki/Flow_(psychology)
The reason straight jackets lead to poor usability is that human minds are fairly diverse in the way they solve problems, and rigid constraints imposed by a tool are likely to trip up your thinking.
There is a tension between flow and feedback, but its not clear at all that one dominates the other as you've stated.
Ideally, we could achieve flow without sacrificing semantic feedback (or vice versa). It is definitely a worthy goal
Is it anti-human to be able to interact with the code, and explore what it actually does? A run-time type error is one that has real example data.
The assumption we can write perfect code immediately and easily, or that a type analysis can just guide us through, is not the case. And it's breaks down even worse, when 99% of your code is interacting with systems outside of the scope of the typing system. (database servers, client-side browsers, network connections, file-systems). Dynamically typed language are good fits, when the code is mostly glue-code between multiple systems outside of the scope of any type analysis.
But let's not equate functional programming with static typing.
>in languages that provide a strong theoretic framework for reasoning about code.
You act like many of us are ever writing complicated algorithms. We're not. We're writing simple algorithms that deal with complex structures of information. And in the few cases where the algorithms get so complicated you want your invariants to be formally proven, anything less than a full blown theory prover, will be insuffient anyway.
While dynamic procedural languages have their problems, I think they work for the most part, hence their success Vis-à-vis functional languages. As far as the issue of scale, that's why modular constructs as added to manage large code bases.
The adoption of programming languages is more influenced by economic factors than informed by solid engineering.
We can debate this ad infinitum and probably wont come to an agreement.
I would argue that the "anything goes" dynamic procedural languages ( like Ruby, Python ) are far more anti-human, in that human ability to reason about large masses of code scales orders of magnitude worse than in languages that provide a strong theoretic framework for reasoning about code.