Improve lecture flow and reorganize content

This commit includes pedagogical improvements across three lectures:

**numba.md:**
- Improve sentence flow with better transitions
- Change Wikipedia multithreading link to internal reference
- Add "(multithreading)=" label to Multithreaded Loops section
- Remove "Numba will be a key part of our lectures..." sentence
- Add transition phrase "Beyond speed gains from compilation"
- Clarify NumPy arrays "which have well-defined types"
- Change "For example" to "Notably" for better flow
- Add "Conversely" transition for prange vs range comparison

**numpy.md:**
- Add "### Basics" subheading for better organization
- Emphasize "flat" array concept with bold formatting
- Improve shape attribute explanation with inline comments
- Remove np.asarray vs np.array comparison examples
- Remove np.genfromtxt reference, keep only np.loadtxt
- Remove redundant note about zero-based indices
- Improve searchsorted() description formatting
- Remove redundant NumPy function examples (np.sum, np.mean)
- Simplify matrix multiplication section (remove old Python version notes)
- Simplify @ operator examples, remove redundant demonstrations
- Remove manual for-loop equivalent of broadcasting
- Remove higher-dimensional broadcasting code examples
- Remove higher-dimensional ValueError example
- Add "### Mutability" subheading and improve organization
- Change "Vectorized Functions" to "Universal Functions" heading
- Emphasize terminology with bold: **vectorized functions**, **ufuncs**, **universal functions**
- Add note about JAX's np.vectorize
- Remove "Speed Comparisons" section (moved to numpy_vs_numba_vs_jax.md)
- Remove "Implicit Multithreading in NumPy" section (moved to numpy_vs_numba_vs_jax.md)

**numpy_vs_numba_vs_jax.md:**
- Change title from "Parallelization" to "NumPy vs Numba vs JAX"
- Add jax to pip install command
- Add missing imports: random, mpl_toolkits.mplot3d, matplotlib.cm
- Add "### Speed Comparisons" section (moved from numpy.md)
- Add "### Vectorization vs Loops" section (moved from numpy.md)
- Add "### Universal Functions" section (moved from numpy.md)
- Add "### Implicit Multithreading in NumPy" section (moved from numpy.md)
- Change "some examples" to "an example" in multithreading description

🤖 Generated with [Claude Code](https://claude.com/claude-code)

Co-Authored-By: Claude <noreply@anthropic.com>

Author: jstac

lectures/numba.md

@@ -67,9 +67,7 @@ The key idea is to compile functions to native machine code instructions on the
 
 When it succeeds, the compiled code is extremely fast.
 
-Numba is specifically designed for numerical work and can also do other tricks such as [multithreading](https://en.wikipedia.org/wiki/Multithreading_%28computer_architecture%29).
-
-Numba will be a key part of our lectures --- especially those lectures involving dynamic programming.
+Beyond speed gains from compilation, Numba is specifically designed for numerical work and can also do other tricks such as {ref}`multithreading`.
 
 This lecture introduces the main ideas.
 
@@ -244,7 +242,7 @@ Successful type inference is a key part of JIT compilation.
 
 As you can imagine, inferring types is easier for simple Python objects (e.g., simple scalar data types such as floats and integers).
 
-Numba also plays well with NumPy arrays.
+Numba also plays well with NumPy arrays, which have well-defined types.
 
 In an ideal setting, Numba can infer all necessary type information.
 
@@ -297,7 +295,7 @@ As mentioned above, at present Numba can only compile a subset of Python.
 
 However, that subset is ever expanding.
 
-For example, Numba is now quite effective at compiling classes.
+Notably, Numba is now quite effective at compiling classes.
 
 If a class is successfully compiled, then its methods act as JIT-compiled
 functions.
@@ -447,6 +445,7 @@ function.
 
 When Numba compiles machine code for functions, it treats global variables as constants to ensure type stability.
 
+(multithreading)=
 ## Multithreaded Loops in Numba
 
 In addition to JIT compilation, Numba provides powerful support for parallel computing on CPUs.
@@ -603,7 +602,7 @@ independence fails, and this is why we use ordinary `range` instead of `prange`.
 When you see us using `prange` in later lectures, it is because the
 independence of tasks holds true.
 
-When you see us using ordinary `range` in a jitted function, it is either because the speed gain from parallelization is small or because independence fails.
+Conversely, when you see us using ordinary `range` in a jitted function, it is either because the speed gain from parallelization is small or because independence fails.
 
 ## Exercises