Improve decoding performance by 40% by preallocating arrays of known length

[adriancable]

Currently the generated .fromObject methods contain code like this:

                if (m.properties && m.properties.length) {
                    d.properties = [];
                    for (var j = 0; j < m.properties.length; ++j) {
                        d.properties[j] = $root.firebase.DeviceProperty.toObject(m.properties[j], o);
                    }
                }

This is bad, because you're doing m.properties.length allocations. We can do better since the size of the array is known:

                if (m.properties && m.properties.length) {
                    d.properties = Array(m.properties.length);
                    for (var j = 0; j < m.properties.length; ++j) {
                        d.properties[j] = $root.firebase.DeviceProperty.toObject(m.properties[j], o);
                    }
                }

This PR improves decode performance on the benchmark by around 40% (similar gains in the real world, too).

Before:

benchmarking decoding performance ...

protobuf.js (reflect) x 1,539,774 ops/sec ±4.99% (75 runs sampled)

After:

benchmarking decoding performance ...

protobuf.js (reflect) x 2,186,209 ops/sec ±1.41% (90 runs sampled)