Storage layer preferred TIMESTAMP_PROVIDER not used for initial setup from local file

Nobody has claimed this yet.

## Description

For confirmed bugs, please report:
- Version: tested with 0.3.1 and 0.4.0 but the code is still percent in master
- Storage Backend: tested with HBase but affects all

When initializing a new graph where the storage layer has a preferred `TIMESTAMP_PROVIDER` value different than the default, the initial settings are stored using the wrong timestamp resolution.

I found this bug trying to delete some GLOBAL_OFFLINE settings in a deployment using HBase as the storage layer. HBase was set to use _milliseconds_ timestamps but the graph settings specified in the `.properties` file were being saved using _microseconds_, which is the default value for `graph.timestamps`.

### Expected Behavior
All configuration values should be set using the preferred storage layer timestamp resolution when available.

### Current Behavior
Initial configuration values always use the default timestamp provider resolution.

### Steps to Reproduce
With HBase running locally using milliseconds timestamp resolution (the default).
1. Create a new graph with: `JanusGraphFactory.build().set("storage. backend","hbase").open()`
2. List the content of the `janusgraph` table with `scan "janusgraph"`

_Expected:_ timestamps are milliseconds values.
_Actual:_ timestamps are microseconds values.

Start with the JanusGraphFactory.build().set(...).open() initialization path and inspect how settings from the .properties file are written before the storage backend's preferred TIMESTAMP_PROVIDER is applied. Reproduce with HBase and scan the janusgraph table; done means initial configuration timestamps use the backend's preferred resolution rather than the default.

Written by the indexing model from the issue text.