Also cover loop method.

It will make further testing much easier

If at least once certificate has been downloaded from KeDiFa it shall never
use again the fall-back, as otherwise it would result with a problem, that
next unsuccessful download from KeDiFa would result replacement with
fall-back.

In order to do so state file is introduced keeping list of overridden
certificates. As now there is critical path regarding fetching certificates,
the lock is created to avoid concurrent updates.

Fix one condition.

In order for further development and features create mixin.

Features:

 * by default runs with 60s sleep
 * allows to have master, updateable, certificate, which is used in case if
   specific certificate is not available

Having PRIMARY KEY on certificate.id is to strict -- as real uniques is
required on id + reference in certificate table.

capturer does not work in some of tests environment.

In some places it is not working, and anyway KeDiFa is used in pinned
versions environment mostly.

Provided tools are kedifa and kedifa-getter.

kedifa is a server to PUT and GET sensitive information, like SSL keys and
certificates.

kedifa-getter is a client to this server.

As both are closely related to caucase, they allow to use information from
caucase, like CA Certificate, to validate each other.

Caucase is also used to generate certificates for kedifa-getter used to
authenticate to kedifa.

Extracted important points of development of the inital version:

 * kedifa and kedifa-getter has been implemented
 * TODOs list is kept for future improvements
 * IPv6 and SSL-only support came
 * API has been docstring documented
 * PUTting information is based on query string key authorisation
 * GETting information requires SSL authentication
 * only correct keys are stored in KeDiFa database
 * certificates are served orderd by theirs submission date
 * kedifa-csr has been implemented, and dropped, as started to become openssl
   req implementation
 * caucase.http has been used as base for wsgiref approach
 * caucase.utils has been used for certificate management
 * argparse has been used for command line arguments
 * time comparison has been done in python, instead of SQLite
 * reloading, in caucase way, has been implemented
 * CRLs are in-app checked only, as pythons implementation does not allow
   proper reloads
 * in critical places code raises instead of returning False, in order to
   disallow ignoring result value
 * ids to store data has to be reserved

Use versioneer. Add basic README. Create structure.