[docdb] Add a limit on number of outstanding tablet splits

[bmatican]

Porting point c) from #6111 (comment)

Once we have #5647, we will have the info in the master, of outstanding tablet splits, so we should also be able to limit how many tablets are still being split, so we could reliably add a limit for this. This would help to minimize the amount of work in parallel in a cluster with many tablets that will end up being split.

[bmatican]

One thing we've discussed offline, is that we can do this probably easier on the TS side, to make sure we don't send extra info to the master, about new tablets to split, if we have not finished compacting all the currently being split tablets.

Notes from @ttyusupov

if (!tablet || !tablet->metadata() ||
    tablet->table_type() == TableType::TRANSACTION_STATUS_TABLE_TYPE ||
    // TODO(tsplit): Tablet splitting for colocated tables is not supported.
    tablet->metadata()->colocated() ||
    tablet->metadata()->tablet_data_state() != tablet::TabletDataState::TABLET_DATA_READY ||
    tablet->GetCurrentVersionSstFilesSize() < split_size_threshold ||
    // TODO(tsplit): We don't split not yet fully compacted post-split tablets for now, since
    // detecting effective middle key and tablet size for such tablets is not yet implemented.
    tablet->MightStillHaveParentDataAfterSplit()) {
  VLOG_WITH_FUNC(3) << Format(
      "Skipping tablet: $0, data state: $1, SST files size: $2, has key bounds: $3, has been "
      "fully compacted: $4",
      tablet->tablet_id(),
      tablet->metadata() ? AsString(tablet->metadata()->tablet_data_state()) : "NONE",
      tablet->GetCurrentVersionSstFilesSize(), tablet->doc_db().key_bounds->IsInitialized(),
      tablet->metadata()->has_been_fully_compacted());
  continue;
}

To not just skip tablets with MightStillHaveParentDataAfterSplit , but if we have them - return from this function, so we don’t trigger new split

[bmatican]

was discussing with @robertsami , that if we move the control of when to do splits, to the master, then this task will be even easier

so @mikhpolitov let's hold off on this a bit, until Rob wraps up the master side work then!