You can get into a situation where multiple disconnect events happen
w/o connection events.  This makes it look like the code cannot count
connected channels.  Only change the counts when we have a change from
the current state.

Introduce fixed sized buffers and split the iteration of the channel
list into stages.  This bounds the memory use of the each connected
client.

This allows disconnected channel lists to be extracted from the edc.

We can easily handle floats, int16, int32, so do so.

Turned src/drv/param.c src/drv/crc.c into a helper library.
Modified daqd and standalone_edc to consume this as a library.

Removed more now unused code from standalone_edc.

omx_xmit should run  after local_dc, so it does not need all the code
to wait for individual models and combine/concentrate the data.

during fast ADC testing.

This is done mainly to reduce IPC traffic of WD signals from SUS IOP models.
IPC rate is denoted in model Parameter block as ipc_rate=[desired rate], where
desired rate < model rate and >= 2048 in usual 2^n values.

   - Add specific FE_RATE and IPC_RATE to generated C code file instead of a bunch of ifdefs.
   - Remove old /proc stuff from Makefile generation.

Note that the max IPC rate is still 64K.

TDS timing available.

  - Prematurely ran clang, so unfortunately looks like lots of diffs.
  - Added module initialization for card model numbers being passed to user apps.
    - Without this, user model could mistake a zero as an ADC if IOP does not have the
      right number of cards, such as in a test stand.
  - Added 2msec delay before starting core locked thread to allow time for printk
    buffers to be flushed.

removed from fm10Gen.c.

using controllerIop.c.  Once thoroughly tested, should be able to
deprecate controllerTS.c.

Simply ran clang-format on new mapVirtual.c and time_slave_io.c files that support time slave via Dolphin.

This change also include Keith's changes for latest Dolphin driver code.

For now require edc ini entries to be floats @ 16Hz.

fe_stream_check_edcu_nds.
Given a timespan (w/o gaps) it checks:

 * That there is a consistent number of (*)EDCU_CHAN_(CONN|CNT|NOCON)
values.
  * There can be no non-connected channels for this test
 * It then samples a number of channels that were generated by a
 synthetic generator and verifies they have the expected values.
  * To deal with delays in epics changes being propagated, the comparison
 looks at the current second and the previous second and allows a blending
 of their values.