to transient objects with attributes unset would leak NEVER_SET,
and negated_contains_or_equals would do so for any transient
object as the comparison used only the committed value.
Repaired the NEVER_SET cases, fixes #3371, and also made
negated_contains_or_equals() use state_attr_by_column() just
like a non-negated comparison, fixes #3374

- start writing docs

sqlalchemy/orm, sqlalchemy/event, sqlalchemy/testing

to get all flake8 passing

implicitly initialized to None via first access; this action,
which has always resulted in a population of the attribute,
now emits an attribute event just like any other attribute set
operation and generates the same kind of history as one.  Additionally,
many mapper internal operations will no longer implicitly generate
these "None" values when various never-set attributes are checked.
These are subtle behavioral fixes to attribute mechanics which provide
a better solution to the problem of :ticket:`3060`, which also
involves recognition of attributes explicitly set to ``None``
vs. attributes that were never set.
fixes #3061

scenario, where an INSERT/DELETE can be turned into an UPDATE.
In this situation, a many-to-one relationship set to None, or
in some cases a scalar attribute set to None, may not be detected
as a net change in value, and therefore the UPDATE would not reset
what was on the previous row.   This is due to some as-yet
unresovled side effects of the way attribute history works in terms
of implicitly assuming None isn't really a "change" for a previously
un-set attribute.  See also :ticket:`3061`. fixes #3060

of cascading a natural primary key update will succeed
even if the key is composite and only some of the
attributes have changed.
[ticket:2665]