Add the TS.ADD arguments with the per-call ON_DUPLICATE override, and
the TS.INCRBY/TS.DECRBY arguments with TIMESTAMP, on the shared series
options. The base serializer now places TIMESTAMP first and
ON_DUPLICATE after DUPLICATE_POLICY, keeping LABELS last.

Co-authored-by: HwangRock <peter9244@catholic.ac.kr>

TsMAddValue is one TS.MADD sample. of(key, timestamp, value) requires a
non-negative timestamp; of(key, value) sends * so that the server clock
sets the timestamp.

Co-authored-by: HwangRock <peter9244@catholic.ac.kr>

TS.MADD replies with an array that holds an error element for each
rejected sample. ErrorTolerantIntegerListOutput turns those elements
into null and ErrorTolerantIntegerValueListOutput into Value.empty(),
reading the reply to the end; an error reply instead of the array still
fails the command.

Co-authored-by: HwangRock <peter9244@catholic.ac.kr>

Add TS.ADD, TS.MADD, TS.INCRBY and TS.DECRBY to every API flavor. The
TS.ADD overloads without a timestamp send * (server clock); their
Javadoc warns that a long passed as the value compiles silently. TS.MADD
returns null for each rejected sample (Value.empty() on the reactive
API, List<Long?> on Kotlin), so the reactive tsMAdd and the builder
method tsMAddValues are registered as known deviations.

Co-authored-by: HwangRock <peter9244@catholic.ac.kr>

The advanced cluster API sends TS.MADD with keys in several slots as one
command per slot, in parallel, and puts the results back in the caller's
order. Samples are partitioned by position, because one key can appear
several times. A slot whose command fails fails the whole call, as with
MGET. The split is not atomic across slots.

Add the writes section to the TimeSeries user guide: TS.ADD with and
without a timestamp and the long-to-double pitfall, IGNORE (only with
duplicate policy LAST), TS.MADD partial failure and the per-slot split
in Redis Cluster, and TS.INCRBY/TS.DECRBY.

The Java long-to-double warning does not apply in Kotlin, which does
not widen a Long to a Double. Drop it from tsAdd(key, value) and
tsAdd(key, value, args), and name the overloads that set the timestamp
as tsAdd(key, timestamp, value) and tsAdd(key, timestamp, value, args).

The slot split happens in the Redis Cluster API only. Name it in the
tsMAdd documentation of all six flavors, and state in the class
documentation of both node-selection interfaces that tsMAdd runs
unchanged on every selected node, so its keys must hash to slots that
node serves.