seed in order to re-run with the same RNG, also initializing the RNG
with a tsc based seen by default. Updating some language around seeding the noise part.

delay units now delay some amount during shutdown.
the delay during shutdown is configured as a separate value from the startup delay

Previous code assumed link was top-level, dropping any part of the reference after the second slash.
This led to a link to the wrong subsystem.

Improved error message when linked subsystem is not found in library.

Note that the search code doesn't verify the full path of the link.  If there were another part of the same name
but with different parent subsystems, the RCG would happily link to that instead.  This problem is *not* fixed by this patch.

$bdroot is a reference to the current mdl file.  It's used only in newer (>=2017)
matlab versions.

post process script changed to replace $bdroot with current file name, rather than truncated system name.
Also, a single typo was fixed.

Cherry picked from master

Six channels are created by the EDCU.  Only three were sent to both EPICS and the data stream.
 * Update the standalone_edcu to export the UPTIME, GPS, and DATA_RATE_KB_PER_S values in EPICS as well.
 * Minor formatting changes as per clang-format run.

Closes #332

When waiting for testpoint manager to load.

Several steps were taken to shore up synchronization of awgtpman to the IOP.  These steps do not directly address the crash in issue in #377, but by
blocking execution until the model is fully running, they may prevent the crash from ever happening.

1. Ignore negative card counts from the IOP.  This obviously indicates the IOP isn't loaded.
2. Block execution of a seemlingly loaded IOP until cycle counter is seen to change.  The cycle counter doesn't start updating until the
IOP is fully loaded.
3. Allow up to a minute for an IOP to fully load, then exit nicely.

Changed from using an environment variable to using a function interface for setting sdf_file_loaded

This bypasses the frustrating cases where a snap file (particularly safe.snap) doesn't set the right BURT_RESTORE variable.

Create chans/ipc/<IFO>.ipc if missing during build
Create chans/tmp if missing during installation

Also, change defaults for DACKILL_PANIC variables to NORMAL, and BPTIME variables to 600 sec.

 * Remove the prefix from the parsing of the req file.
 * Add a target to build the sequencer to the cmake, for testing purposes.

Closes #369

We'd dropped an old build flag that was used to determine the element element size for test point numbers.

The testpoint size reverted from 4 bytes to t2, but only in the RPC definition.  As a result, only the first value in an RPC array of testpoints was in the correct position.  Only one testpoint could be specified correctly per test.

I've checked the rest of the code to make sure this was the only remaining use of the build flags.

Fast ADCs were reading only one point per cycle.  The ADC read loop limit was being reset to 1 for slow ADCs but not set back 1 for fast ADCs.

the 4-value dropouts seen on h1ioplsc0 fast adc channels ought to be fixed by this change.

Except <br/> tags are replaced with newlines.

rcg: Moved sequence check into /src/python dir.  Improved sequence error message.  Fixed some typos and formatting issues.

test is implemented as a python script in test/python_modules/sequence/

test compares the calculation sequence to the calculation graph.

The test is run during compilation after parsing but before building.  By default, a failure will kill compilation, but
if IGNORE_SEQUENCE_ERRORS=1 in the environment, only a warning will be issued and compilation will continue.

Two out-of-order bugs are fixed.

1. Ordering simply counts how many inputs of a part of visited. When that number matches the total inputs for the part, the part is visited,
but inside a subystem, delays were counted twice.  Once on the initial pass looking for signal sources, and again as a normal part.

Parts that took input directly from delays could appear early in the sequence.

This change skips following links form delays as a normal part.

2. If a data source, including subsystem inputs, constants, delays, grounds were connected directly to a subsystem output, then
parts immediately connected from the output would be added to the subsystem sequence, though they were at a higher level.

This bug could cause these parts to be sequenced out of order.

Change is to check whether a part leading form a data source is in the subsystem.  This check is already included when following from normal (non-input) parts.