Fixes #774

* rounds nanoseconds boundaries to milliseconds

* convert also store query

moving loki-canary project under loki repo

* Added tail length limit to limit duration of live tailing of logs to 1 hour

* Some code refactoring suggested from PR review for live tailing duration limit

* Fixed error messages in live tailing of logs

* Fixed lint errors

* add benchmark for storage queries
* improve iterator to load only on next
* fix memory retained by lazy chunks
* reverse backward lazy iterator

This commit rolls back the changes to the Makefile and build/Dockerfile
that caused CGO_ENABLED=1 to be present during some builds. This commit
causes the journal support to be disabled in any build produced by make.

Journal support can still be enabled in a manual build:

  go build -o cmd/promtail/promtail cmd/promtail

cleanup rake warnings, bump version

Support for reading systemd journal entries has been added. 
promtail will look for a job in scrape_configs with a journal key 
to activate the journal target. 

If GOOS=linux and CGO_ENABLED=1, promtail will now require 
libsystemd headers to be available for building. If GOOS is not 
linux or CGO_ENABLED is not 1, journal support will be unavailable
and a log message will be printed warning the user that their config 
file has journal tailing configured without it being built into promtail. 

See docs/promtail-examples.md for a concrete example of 
using journal support. 

Other structural changes made: 

  1. Ability for checking if scrape.Config.ServiceDiscoveryConfig is 
     non-zero has been added. 

     This was chosen over making ServiceDiscoveryConfig a pointer
     as yaml.v2 cannot parse an inline struct into a pointer value. 

  2. Updated pipeline logger component name to journal_pipeline and
     file_pipeline for JournalTargetManager and FileTargetManager
     respectively.

  3. The positions file will now store positions as strings instead of 
     integers. Existing positions will be read in properly but written out 
     as strings the next time the positions file is saved. This is done to 
     be able to store the journal cursor, which is a string. The positions 
     API has been updated to support reading in the old integer values 
     and the new string values. 

Fixes #749

Signed-off-by: Brian Candler <b.candler@pobox.com>

Reduces background chatter which in turn can trigger more log writes

Signed-off-by: Brian Candler <b.candler@pobox.com>

* feat(logcli): quiet mode

Adds a quiet mode (-q / --quiet) to suppress the debug messages to stderr

* feat(logcli): output modes

Adds two alternative output modes (-o / --output)

- raw: emits the line as parsed
- jsonl: emits the line plus all known metadata as JSONL (JSON Line)

Usage: -o [default, raw, jsonl]

* feat(logcli): quiet tailing mode

* feat(logcli): output modes while tailing

Supports the three different output modes in tail mode as well

* feat(logcli): print labels in jsonl tail mode

* refactor(logcli): clean up entry printing

Moves the entry printing into a standardized interface, implements this three
times:

- default (human readable)
- jsonl (for scripts)
- raw ('as is')

* feat(logcli): query from absolute timestamp

Adds a new flag (`--from`) as to complement `--since`.
While since subtracts a relative duration from the current time, from allows to
specify the absolute start of the lookback window.

Note: `--from` takes precedence over `--since`, but only if set.

* fix(logcli): use RFC3339Nano for -from

To comply with Prometheus, this flag now honors Nanoseconds as well (was using
RFC3339 so far)

use strconf.FormatFloat instead of fmt.Sprintf for converting floats to strings, this way we can eliminate non significant trailing zeros such that the float value 1 would be "1" as a string instead of "1.000000"

canary will suspend all operations on SIGINT but not exit, allowing you to shutdown the canary without it being restarted by docker/kubernetes
SIGTERM will shutdown everything and end the process

* Query label values and names are now fetched from the store.

A time range is now required by the /api/prom/label with a sane default (6 hours from now).

* fix http querystring and update doc

* update vendor

* rebased

* move Dockerfile multi-arch to build and ignore that folder

* docker-driver fix

* fix(loki): pass missing config error to user

Missing config errors are handled at the library level. Our own check mitigated
this and causes loki to SEGFAULT later on

* feat(loki): default config file in container

The container provides a default config file. Use it by default

* Revert "fix(loki): pass missing config error to user"

This reverts commit b2744fcf427b2b386dce5e999c855ae5c899e4e2, because loki it
assumed loki was incapable of running without config, which is not the case.

Adds an entirely new Dockerfile to the repository root which is capable of:

- building promtail and loki from the same image
- based on `scratch`, so that the final image can be executed on every supported
  GOARCH

Please note that this Dockerfile should always be built using
BuildKit (`buildctl` or `DOCKER_BUILDKIT=1` in recent versions) for maximum
performance, as BuildKit's DAG allows for smart skipping of uneeded stages.

These changes were proposed in #659