Db cleanup

[Christian-B]

several Db cleanups.

avoids running DDL every time file is opened. Like SpiNNakerManchester/SpiNNFrontEndCommon#1235

move caching from each recorder a class variable in neo_buffer_database.py

• for a reset no longer required to ask each Population to cache
• new segement_db method

recorder still gets the block from the earliest segment possible. (to get the start date)

• append_previous_segment( and append_current_segment create the block if needed.
• write_csv_metadata salso tried on each segmenent until one works.

1. For cases where all segments have data this is one less database opening.
2. Small cost for case where earlier segments not recording

merged public method and the private method they wrap

new find_units method to avoid calling get_recording_metadata and then throwing away a lot of data

now also supports where additional variables are added to recording at a reset

Added a test