Don't use Mako templates to generate code

[niknetniko]

I did some profiling of TESTed with the "Star Battle" exercise (in JavaScript). This showed that of the original ~5s execution time, only half was spent executing the actual submission, while half was spent in the Mako templates generating code.

In this PR, I removed the use of Mako to generate code, and used manual string construction instead. Hyperfine results of TESTed with Mako templates:

Benchmark 1: python -m tested.manual
  Time (mean ± σ):      5.561 s ±  0.327 s    [User: 5.417 s, System: 0.208 s]
  Range (min … max):    4.972 s …  6.306 s    10 runs

The same exercise without Mako templates:

Benchmark 1: python -m tested.manual
  Time (mean ± σ):      1.972 s ±  0.276 s    [User: 1.943 s, System: 0.084 s]
  Range (min … max):    1.709 s …  2.380 s    10 runs

So, to summarize:

• This provides a ~2s speedup for all languages, but this will be especially noticable for fast languages (e.g. a Kotlin of Haskell submission spends so long compiling, that this isn't that much of an impact)
• Using normal Python functions revealed some bugs, as we can now use type hints etc.
• Generating code is a bit uglier: in Mako we had to take special care to not put newlines everywhere, while now, you must not forget to put newlines everywhere.

Given that speed is important, I think the trade-off with slightly uglier code generation is worth it.

Fixes #332.

[pdawyndt]

Good job! Has this any impact on the programming language configurations themselves, i.e. do they need to be updated one by one?

[niknetniko]

Good job! Has this any impact on the programming language configurations themselves, i.e. do they need to be updated one by one?

Yes, as this change is purely in the programming languages modules (i.e. there are practically no changes to the core of TESTed). However, I already converted all modules in this PR.