- ensure we do accept **kwargs so any super().__init__(**kwargs) from an
  unexpected MRO path in an indecent maze of multi-inheritance will not crash
- do not call super().__init__(); it is either useless (most probably,
  super() is object in this context) or silently ignore passed kwargs that
  super() (when it's not object) may have been able to understand...

Long story short, kwargs handling in multi-inheritance __init__ methods is
a mess.

- merge raw_post and row_get in a unique raw_verb method,
- merge post_stream() and post() as a unique post(); post_strean is then a
  call to post with the 'stream' parameter set,
- merge get_stream() and get() the same.

so the pytest-postgresql fixture works as expected on non C locale systems.

we almost never return text/plain or text/html, so json is a better default
value.

for debugging purpose.

this will take precedence over the implicit config file scheme.

Related to T1410 and T826

`entry_to_bytes` and its friends were called many times (eg.
entry_to_bytes alone was called 40k times while indexing 500
origins with the metadata indexer), and its use of `isinstance`
used a non-negligible amount of CPU time.

Instead of using `*_to_bytes` function as post-processing on
all bits of data returned by postgresql, this patch tells psycopg2
to use a new `typecast_bytea` function when needed (in `adapt_conn`).
This function deffers the decoding work to psycopg2, which returns a
`memoryview`, which is turned into `bytes`.

to keep the copy_to API bw compatible.

This needs tests!

so thath it does not fail if no payload at all is provided in the request,
eg. for an endpoint which really is (or at least should be) a simple GET...

just to make our life a bit easier...

so one can use this method with default values for some columns. This is used
for example in scheduler.

it is not a 'private' method. Add a _cursor alias for bw compat.

There is no need for it to be declared there: it does not use any closure
and defining a function in a function is expensive!

so that we do have to manage yet another dependecy; the code is small and has
not been modified for years.

that recursively merge 2 (config) dictionaries.
This is meant to be used to simplify a bit the config system of several
systems of the swh platform (related to T1410 and T826).

and raise logging level when loading a config file to info.

This class is meant to be used not only to wrap storage-like db.

Also rewrite the test_logger using pytest-postgresql fixture so
we do not need pifpaf anymore.

Using these 2 negotiate formatters, it is possible to write Flask endpoints
as:

  from flask import Flask
  from negotiate.flask import negotiate
  from swh.core.api import JSONFormatter, MshpackFormatter

  app = Flask()

  @app.route('/somewhere')
  @negotiate(MsgpackFormatter)
  @negotiate(JSONFormatter)
  def somewhere():
      return {'some': 'content'}

See https://github.com/nickstenning/negotiate

prometheus-statsd-exporter uses the datadog format for its tags, so base our
client on their Python code, with the following modifications:

 - Removed python < 3.5 compat code
 - trimmed the imports down to be a single module
 - move to time.monotonic() instead of time.time()
 - adjust some options:
   - drop unix socket connection option
   - add environment variable support for setting the statsd host and
     port (pulled the idea from the main python statsd module)
   - only send timer metrics in milliseconds (that's what
     prometheus-statsd-exporter expects)
   - drop DataDog-specific metric types (that are unsupported in
     prometheus-statsd-exporter)
 - made the tags a dict instead of a list (prometheus-statsd-exporter only
   supports tags with a value, mirroring prometheus)
 - improve unit test coverage
 - documentation cleanup

Related D816?id=2603#inline-4587

The first time, the db will be initialized.
The next time this is called, this will do nothing.