I was a TA for the introductory programming class at Berkeley, CS61a, where we used Structure and Interpretation of Computer Programs to teach Scheme, a statically-scoped dialect of LISP. Frankly, I do not recommend it. Scheme is just another programming language, it is not "the one universal programming language to rule them all and in the darkness bind them".
I spent years in mathematics and CS theory and I am a published theoretician. I took two graduate classes from Karp himself and even solved one of Karp's open theory problems because I thought it was a bonus homework problem in an undergraduate class. I am also a programming languages researcher and a serious C++ dev and am now teaching myself digital circuit design. I wrote an integer user-mode RISC-V 64 chip that worked the first time I got it expressed in verilog. That is, I know what I am doing in both theory and practice.
It sounds like you want to know the "heavy, deep, and real" computer science. Never forget this: the only theory worth learning is that which comes from real problems. The theoreticians will tell you otherwise, but do not listen to it. Always think about reality first, theory second. Therefore: first just learn how to code like you know how to breathe. Learn one editor and one language/ecosystem really well. I use GNU emacs and I write most everything in C++.
The first language I recommend is RISC-V assembly language. Then learn C and configure gcc to emit the intermediate assembly interleaved with comments showing the C code that it came from. Practice writing C, predicting the assembly that will come out, and then seeing what actually comes out. Practice stepping through your programs using the debugger. Practice profiling them using gprof. Write lots of little programs that do something mundane and useful, write tests and documentation for them, and release them as open source. See if you can implement one program that someone else finds useful. One way to get started with an established open source project is offer to write tests for them. The best way to learn fancy algorithms is to implement them.
