Cache the CMakeDeps Jinja2 templates

[fredizzimo]

Changelog: Fix: Speed up the CMakeDeps generation
Docs: omit

• Refer to the issue that supports this Pull Request.
• If the issue has missing info, explain the purpose/use case/pain/need that covers this Pull Request.
• I've read the Contributing guide.
• I've followed the PEP8 style guides for Python code.
• I've opened another PR in the Conan docs repo to the develop branch, documenting this one.

The compilation of the Jinja2 templates is quite slow, and several templates needs to be compiled per dependency, so for larger projects with a lot of dependencies it can be quite slow. This is fixed by caching the templates per class instance. For a medium size project with around 80 packages and dependencies I saw an improvement from 1:10 seconds to 10 seconds with this fix, so it's quite a bit faster.

NOTE: I'm not particularly happy with the implementation itself, but any alternative implementation I can think of need all the template files to be modified. So I decided to go with this non-intrusive implementation. But I can change it if you prefer.

Fixes #13845

[memsharded]

Looks good, thanks!

I have realized that there is a potentially unnecessary slow textwrap.dedent() for every large string, for every file, and this PR also optimizes it, not only jinja.

[memsharded]

Doubled checked, all .template are indeed constant strings, all good.

[memsharded]

NOTE: I'm not particularly happy with the implementation itself, but any alternative implementation I can think of need all the template files to be modified. So I decided to go with this non-intrusive implementation. But I can change it if you prefer.

I was also initially thinking something a bit more explicit, without getattr(type(self)..., but the PR is short, only in one single place instead of multiple files, so I think it has merit, and is good enough to move forward. Thanks again!