Optimize RequestContext construction

[Bastian Kleineidam <calvin@…>]

The current RequestContext constructor pushes a new context variable
onto the stack for each template processor result. This is not
necessary. This patch just updates the top dict entry of the
current context instead of pushing a new one onto the context dict stack.

This might break some apps that pop() template processor contexts after
they have been added, but that would be using an undocumented feature, and its
a bad idea: why add the contexts in the first place if they get removed afterwards.

[Łukasz Rekucki]

Modifying dict[0] isn't a good idea as those most likely would be the builtins that are initialized in BaseContext._reset_dicts, which means they get overshadowed by dict_ passed to the constructor.

Currently context processors overshadow the dict_ (I seen people debug their code for hours before realizing that, a real PITA), so any change to this behavior needs to be documented. A test case to show that no extra dicts are created while the right order is preserved would be good too :).

[FunkyBob]

Given that most of the rest of the dict-ish methods on BaseContext use self.dicts[-1] ... it would make sense for this to, also.

I've tested it locally, and it passes all current tests.

[FunkyBob]

I've created a PR for this - ​https://github.com/django/django/pull/1527

I put all the context processor data into its own stack dict, on the off chance someone relies on the dict they pass to RequestContext not being altered.

[Anssi Kääriäinen <akaariai@…>]

In 8d473b2c54035cbcd3aacef0cb83a9769cd05ad3:

Fixed #7116 -- Optimize RequestContext construction