Log each Sidekiq line once, with the job's ID, and write logged Hashes at the top level - #17

The Sidekiq logger put the live Sidekiq::Context hash into each log
struct. SemanticLogger formats on its appender thread, so for a fast job
it could still be iterating that hash when Sidekiq's JobLogger added
:elapsed after the job finished. The resulting "can't add a new key into
hash during iteration" error was raised after the job had succeeded, so
Sidekiq marked it as failed and retried it.

Fixes #14

Sidekiq's own log lines carry the job ID and class in ctx, but lines the
job itself writes with Rails.logger don't, so there was no way to pull
every line for one job run out of the logs. Until the broadcast fix,
the duplicate Sidekiq copy of each line was the only one with a job ID.

A Sidekiq server middleware, first in the chain, now pushes the job's
jid and class as SemanticLogger named tags for the job's duration, and
the formatter writes them to ctx unless the line already has one.
ActiveJob jobs show their own class rather than the Sidekiq wrapper,
matching Sidekiq's context.

The logger passed a Hash to SemanticLogger as `payload: hash`, which
SemanticLogger stores as {payload: hash}, so Rails.logger.info(src:
"rails", evt: "rate_limit", ...) was written as
{"payload":{"src":"rails",...},"src":"app","evt":"log"}. Fields that
log searches and metric filters expect at the top level were nested
under "payload", and the line's own src and evt were overridden. Hashes
are now passed as the payload itself; structs are unchanged.

A logged Hash may set its own src and evt. Its ts and lvl are reported
as LogStruct errors and skipped, since the line's timestamp and level
come from the logger. During a Sidekiq job, a Hash's own ctx keeps its
keys and gains the job's jid and class.

Sidekiq's default error handler logged three warn lines for each failed
job: the job hash as a JSON string inside msg (with the job's
arguments, which LogStruct's filters can't reach there), the error, and
the backtrace. LogStruct replaces it with one structured line (src
sidekiq, evt error) with the error class, message, backtrace, jid, job
class, queue and retry count, and no arguments. ActiveJob jobs' errors
are left to the ActiveJob integration, which already logs them. Error
handlers the app added stay.