A) If you're writing high-performance code DON'T you _will_ hit bottlenecks which are non-trivial to the language unless you're a guru. If you're a scientist you're going to have to work to get to this level.
Apart2) Python is excellent for saying "perform analysis A with variables B,C,D,E,F on X,Y,Z 30 times using this random seed library". What Python is not excellent at is saying, "I have 20GB of raw doubles and I need to push these through non-euclidian algebra as fast as a modern CPU can cope". The biggest surprise is that modern Python really can do both, it just struggles a lot with the latter over the former. IPython and JuPyter notebooks as well are amazing for the former!
B) For the love of whatever deity you worship. PLEASE DOCUMENT YOUR DEPENDENCIES CORRECTLY. This is non-trivial and doesn't mean "Works on CentOS6". I mean at least tested on Python x.y with depA version 1.2.3. Python isn't as bad a npm, but it's getting there...
Bpart2) Unless you're starting out, _avoid_conda_. Installing/maintaining things through this is not nice. This is the sort of system that makes your local sysadmin groan when you come to him and say "it's easy because it has a UI". Behind the scenes it makes so many assumptions that are great for people starting out, but are often painfully wrong for people looking to run code at hyper-scale on clusters. I recommend investing _some_ time in learning how python packages work (not much!, packaging, is not coding). But as a result you will learn or employ a lot of skills which will help you with larger python codebases. There are large performant python packages/tools out-there and they all tend to leverage advanced features for good reasons.
C) Never! re-invent the wheel unless you know you explicitly need to. And avoiding pulling in a dependency (unless the dependency is _huge_ is not always the correct answer).
D) Ask an expert. It's fine to acknowledge you don't know how to use the tool that is modern python. It's a swiss-knife with a shotgun attached pointed at your face. When it goes wrong it will come back to bite, but when you get it correct it's so efficient you will marvel at how much you can do with just 5/6 lines of code.
E) Have fun, just try things and see if they don't work or work really well, python is great for avoiding too much boilerplate for mid-scale projects :)