Reworked errors in template syntax and overall more validation, caching by default

[michalpokusa]

Continuation of #3

💥 Breaking changes:

• Language enum, language parameter and Markdown/XML support was removed

⭐ Added:

• Token class for keeping track of tokens postion within template
• TemplateNotFoundError and TemplateSyntaxError errors that are raised instead of SyntaxError and OSError

🗑️ Deleted:

• Support for XML and Markdown, the code was over 1000 lines long, which raised error with Pylint, also, there was no really any practical use for them. It in the future an issue about that is created, then a support for custom escaping function might be a good idea, as I can imagine someone wanting to generate a huge JSON output using templating

🛠️ Updated/Changed:

• While being processed, the template is not being replaced after processing each part, instead processed part is stored as offset, which enables referencing whole template even after processing certain parts, also less asign operations
• Instead of SyntaxError and OSError, custom TemplateNotFoundError and TemplateSyntaxError are raised
• Error are raised in more cases, e.g.
  • when circular extends are detected
  • when there are any tokens between blocks in extending template
  • when else, elif, empty, endwhile or similar is used without previous for, while, if
  • when extends/endblock token is found in top-level template (e.g. when was inserted from child's block)
• When error regarding template syntax is raised, it prints the token and surrounding lines for easy determining the problem source (screenshot below)
• render_... functions now accept cache parameter (True by default) that saves the template the first time the function is ran and then reuses it, even with different contexts, this should improve performance in most use cases as previously one had to create a class instance, store it in global scope and then call its method. Although that still works, now there is a more readable alternative.

🏗️ Refactor:

• Some minor function/variable renaming/moving for better readability

This is example how the new error points to token that caused it:

---

Performance 🏎️:

When it comes to performance, from my testing on somewhat complicated template with 40+ tokens, multiple includes and levels of extending I noticed ~15% more time is needed, of course it only applies to creating the template itself, not using it's render methods. The additional pass statements seem to make no difference on the speed of the function, even when there are hundreds of them.

On a simple template that does not require reading a file from disk I did not see any significant increase in time needed, sometimes the new version was even faster that current one. This might be explaing by the fact thst new version does not replace the value of template (potentially a really big string), but only moves the offset while keeping the template unchanged for the most part, and that might be simply faster, even considering the need to create Token instances.

It seems like reading templates from disk is the thing that is the biggest bottleneck.

Partly because of this I changed how caching works.
Before, all render_ functions did not cache the template, but because they are more handy to use than manually creating a Template instance, now they cache created template by default, which can be disabled using cache=False argument.

It also makes the code cleaner, as it is no longer need to store a template in global scope.

[michalpokusa]

All template syntax error that I was able to think of should be catched, if I missed something please point to that and I will update.

Thanks in advance for testing/review.

[FoamyGuy]

This looks good to me.

I tested the modified examples successfully on a Feather S3 TFT. I also tested the template example from the httpserver repo examles/ dir.