Prefix the retry limit property name with core_workload.

[core] Incorporate CR Feedback

Prefix the retry limit property with "core_workload".

The default behavior is NO RETRY, which is the current behavior.

But optionally user can enable retry if they are working with a DB backend that is better tolerated by allowing a few retries.

- Remove trailing whitespaces,
- Break longlines to below 120 chars
- Right indentation.

This is to prepare for the next commit that will add retry logic to insert to make the load process more fault tolerant.

Define constants for the property strings.

- Move the doc (code comment) to workload_template
- Minor changes in propery parsing logic and log to System.Err;

[core] Incoporate CR comment

Missed one comment before.

And make it configurable.

Specifically:

1. Remove trailing spaces,
2. Replace TAB with spaces,
3. Break lines > 80 characters
4. Misc code alignment.

No functional code change. use ?w=1 on the URL to view non-whitespace changes.

When dataintegrity and readallfields are both true, pass the full field
set to db.read() so that the field values can be verified.

Fixes #496

Changed measurement latency to actual latency time of the verifyRow step
and now recording Status codes for verify operations.

Fixes #479

Also fix a few README formatting issues.

        mvn org.apache.rat:apache-rat-plugin:0.11:check -D rat.numUnapprovedLicenses=5000

1. Use LinkedList rather than ArrayList for the measurement array in RAW. Update size estimation in documentation in workload_template.

2. Add a new composite measurement type which combines HdrHistogram and RAW.

3. Allow user to optionally disable output of summary stats from the raw measurement class.

Replaces numeric status codes with a canonical set of Status objects,
each with a short name and description.
Bindings with more specific errors (e.g., timeouts) return additional
statuses.

This changes the default output from messages like:

   [UPDATE], Return=0, 511

To:

   [UPDATE], Return=OK, 511

Raw datapoint output is quite useful and important for two reasons:
 1) Allows the user to do more thorough statistical analysis on the complete set of the data.
 2) Using timestamps that are logged with raw data points, it becomes much easier to investigate and correlate latency outliers at the backend service.

In OneMeasurementHdrHistogram, now reading hdrhistogram.percentiles
from workload configuration. This property is a comma delimited string
of Integers to use as percentile values.

This implements #437

size for the window so a mask can be used instead of a modulo operation.
This avoids issues when there is overflow of the id integer values to
negative values.

The HdrHistogramReader makes some problematic assumptions on the roles
played by start time, base time and histogram start/end times. This is
to be revisited in a future release of HdrHistogram (see
https://github.com/HdrHistogram/HdrHistogram/issues/69). In the meantime
this change will make YCSB record as expected by the reader:
a. Set histogram log base time (same as start time)
b. Only create the recorder after the log has been setup

final, refactor repeated export code from subclasses.

closes #239.

Origin patch in PR-286.