Implement data retention

[narqo]

There is almost no sense to keep the profiling data for more than N days.

The exact implementation is in TBD. Open questions:

• should this be a general part of cmd/profefe daemon or a dedicated tool for particular storage?
• TODO

[narqo]

"Google-Wide Profiling: A Continuous Profiling Infrastructure for Data Centers" describes an analytical approach for comparing several profiles, based on "profile's variation (entropy)" (see "Stability of profiles"):

We use a single metric, entropy, to measure a given profile’s variation. In short, entropy is a measure of the uncertainty associated with a random variable, which in this case is profile samples.
[..] When we need to identify the changes on the same entries between profiles, we calculate the Manhattan distance of two profiles...

The question is: can a similar approach be used to identify profiles that are no longer needed because the data they contain is redundant comparing to data of other profiles.

[narqo]

Following #92 for storage/clickhouse this can be done via ClickHouse's TTL directive in database's DDL. Refer to docs for MergeTree table engine