* implement browse_next in client

* first attempt to decode custom extension objects, broken!

* fix several small errors

* add 0 as default for max nodes in browse, cleanup

The problem was the missing path and using None
for Int and Bool properties

* xmlexport improve compat with xml schema

* revert changes to _extobj_to_etree, do fix argument member order in other pr

* set default access level of nodes to read

* Revert "set default access level of nodes to read"

This reverts commit c345470e8abc7bfa7594143b22570006b9e4754d.

* set default access level of nodes to read

* Added: handling of Byte, SByte and DateTime convertion from ua to python

* log nodeid in addressspace

* Updated: Removed some duplicate xml parsing.

_parse_value was parsing the child of a node as a constant.
Making it parse a node directly and not its child allows recursive calls
to parse ListOf elements.
_parse_containted_value as been added as a wrapper to obtain former
behavior.
Now ua_type_to_python is simply a wrapper to ua_utils.string_to_val and
_to_bool a wrapper to ua_type_to_python.
All value parsing is performed by ua_utils.string_to_value.
Missing types have been listed as FIXME.

* Added: tests for XML parsing.

Added some tests in tests/tests_xml.py for datatype DateTime (+list),
QualifiedName (+list), ListOfGuid and ListOfExtensionObjects.
Few tweaks in xmlimporter: (i) DateTime is now parsed in string_to_val
and double parsing was causing an error, (ii) single Guid were converted
to UUID while lists were not by default list case.
Another possibility would be either to make the conversion to UUID in xmlparser
or to make the _add_variable_value also recursive but I'm not sure of
other possible consequences.
Finaly it seems according to specs that all DateTime should be timezone
aware so I forced UTC when a timezone was not precised.
Maybe an exception should be raised?

* Added: pytz in requirements

* Added: pytz in travis requirements

* Added: pytz in travis requirements

* Changed values for RequestedLifetimeCount and RequestedMaxKeepAliveCount

Also made them changeable via kwargs

* added comment

* reworked so period can be a CreateSubscriptionParameters

Maybe this works better...

* Split a space dump and cache

dumping and caching really serve different purposes; also cache is not
fully implemented and should only be used in special cases

* Fix dead history methods, docs

Fix history methods and add support for historizing a list of nodes
Add doc strings
pep fixes

* Cache fixes, close session when server stops

fix caching of standard address space
try to suport windows cache file
close session on stop so that all subscriptions get removed (need to
serialize nodes)

* Fix cache file check, fix tests

* Improve lazy loading dict

at least support builtins for cache, otherwise features like
node.delete() don't work

* Update high level historize method, update examples

* let caching and dump/load work at same time

experimental commit for discussion
these changes let caching and dump/load work together

* Revert "let caching and dump/load work at same time"

This reverts commit 0fde446e3f2c105aad7aacc34c243414b7accc4e.

* Remove uneeded import

* Update history tests

* make naming more explicit

* Small name refactor, clarified doc strings

* export to xml based on namespaces

* cleanup of ua_utils.get_nodes_of_namespace

* fix export Enum and custom datatype variables

* fixed extensionobject export again (removed typo)

* Test for import/export EnumValues

* move XmlExport.get_variable_basetype to ua_utils.py

* undo cleanup of get_variable_basetype

Can't check 'not val' because variable nodes with a value of 0.0 don't
get exported.

* do not create Variant with value None when they are supported by protocol

* add get_default_value to uatypes. This will be usefull at several places like opcuamodeler

* set correct default value in high level type creation method

* Made subscribe handle most iterable items 

like dicts, etc

* replaced try/except with isinstance