Logic from Table::MutateRow() will be moved here and this commit
proposes the general structure of this class, for early review.

And unconditionally return an error status for the aggregation
mutations, which we don't currently support.

…quest.

Currently in use only for SetCell.

…anging.

…ion.

- Introduces a RowTransaction class to proxy mutations and record undo
  logs in case we needs to rewind.
- Re-implements the SetCell mutation in the RowTransaction class.
- Introduces all Undo types/actions for SetCell and records them in an
  undo stack.
- For SetCell only and to facilitate testing with only SetCell
  mutations, introduces the use of RowTransaction in doing the mutation.
- clang-format is run on all affected files (which has removed some
  headers, e.t.c.

A test for SetCell is coming up and is not currently included here.

…ulator.

…cted table.

Populate a schema protobuf with the minimal fields and call
Table::Create();

…utation.

There is a crash later on, in the code that sets up the undo log, that
will be fixed in an upcoming commit.

* emulator: mechanical formatting fixes by ci tool.

* emulator: make clang-tidy happy.

* emulator: trivial: fix typo in rollback_test.cc

* emulator: mechanical formatting fixes.

… they revealed in the code.

…e. (#20)

It now has tests well beyond the initial rollback tests, dealing with
all manner of bigtable mutations, so call it mutations_test.cc.

This implements the AddToCell mutation.

* Unit tested and tests passing
* Supports atomic rollback
* Support Sum, Min and Max aggregations for non-negative Int64 numbers
* It is now possible to create an aggregate column family (supporting AddToCell or MergeToCell)

…ixes' into fix_colmnrange_filter

We need it for tests of filters and ReadRow that do not have to or
cannot start up the GRPC network server infrastructure.

…family.

- There is a passing unit test that tests for this (which is fixed
and shows up in this diff).

- The CI test (bigtable_filters_integration_test) which revealed this
bug now passes.