As suggested by Anssi. This has the slightly strange side effect of
passing the expression to Expression.convert_value has the expression
passed back to it, but it allows more complex patterns of expressions.

Refactored compiler SELECT, GROUP BY and ORDER BY generation.
While there, also refactored select_related() implementation
(get_cached_row() and get_klass_info() are now gone!).

Made get_db_converters() method work on expressions instead of
internal_type. This allows the backend converters to target
specific expressions if need be.

Added query.context, this can be used to set per-query state.

Also changed the signature of database converters. They now accept
context as an argument.

This is a reliqua from the early days of the modeltests/regressiontests era.

Complete rework of translating data values from database

Deprecation of SubfieldBase, removal of resolve_columns and
convert_values in favour of a more general converter based approach and
public API Field.from_db_value(). Now works seamlessly with aggregation,
.values() and raw queries.

Thanks to akaariai in particular for extensive advice and inspiration,
also to shaib, manfre and timograham for their reviews.

Signed-off-by: Jason Myers <jason@jasonamyers.com>

* Removed trailing whitespace.
* Added newline to EOF if missing.
* Removed blank lines at EOF.
* Removed some stray tabs.

Squashed commit of the following:

commit 63ddb271a44df389b2c302e421fc17b7f0529755
Author: Aymeric Augustin <aymeric.augustin@m4x.org>
Date:   Sun Sep 29 22:51:00 2013 +0200

    Clarified interactions between atomic and exceptions.

commit 2899ec299228217c876ba3aa4024e523a41c8504
Author: Aymeric Augustin <aymeric.augustin@m4x.org>
Date:   Sun Sep 22 22:45:32 2013 +0200

    Fixed TransactionManagementError in tests.

    Previous commit introduced an additional check to prevent running
    queries in transactions that will be rolled back, which triggered a few
    failures in the tests. In practice using transaction.atomic instead of
    the low-level savepoint APIs was enough to fix the problems.

commit 4a639b059ea80aeb78f7f160a7d4b9f609b9c238
Author: Aymeric Augustin <aymeric.augustin@m4x.org>
Date:   Tue Sep 24 22:24:17 2013 +0200

    Allowed nesting constraint_checks_disabled inside atomic.

    Since MySQL handles transactions loosely, this isn't a problem.

commit 2a4ab1cb6e83391ff7e25d08479e230ca564bfef
Author: Aymeric Augustin <aymeric.augustin@m4x.org>
Date:   Sat Sep 21 18:43:12 2013 +0200

    Prevented running queries in transactions that will be rolled back.

    This avoids a counter-intuitive behavior in an edge case on databases
    with non-atomic transaction semantics.

    It prevents using savepoint_rollback() inside an atomic block without
    calling set_rollback(False) first, which is backwards-incompatible in
    tests.

    Refs #21134.

commit 8e3db393853c7ac64a445b66e57f3620a3fde7b0
Author: Aymeric Augustin <aymeric.augustin@m4x.org>
Date:   Sun Sep 22 22:14:17 2013 +0200

    Replaced manual savepoints by atomic blocks.

    This ensures the rollback flag is handled consistently in internal APIs.

Should be unneeded with Python 2.7 and up.
Added some unicode_literals along the way.