This reverts the changes to Load from commits
  - 67348e40
  - cf334a95

We need a better way for a specific loader to request that it will handle
all files that Load has found.
Refs #10061

The converted text is stored in separate .rst files in the docs directory.
Refs #9562

ITableWorkspace has been added to the list of output workspace types load is expecting to output.

Mainly removed Poco includes and used forward declarations instead.
Unsuprisingly this has affected some algorithms.
Refs #8784

* Remove temp workspace at end of summing.
* Ensure output of load is in the workspace it's supposed to be.

In the case of LoadEventPreNexus is is "EventFilename". It seems that for
everything else is is "Filename" because the variable that holds this is
empty by the time the algorithm comes to run (though the loader is
selected correctly).
I've factored out the method that sets the variable and called it again
from within a method that exec() calls.

Otherwise, when loading sqw, it complains about unknown extension when logging is done in debug mode

Refs #7852

Re #7253

re #6485

Avoids loading everything then summing in separate loop, instead loads then sums in same
loop

Ensure that input into a MultipleFileProperty properly survies a "round
trip", i.e. what you put in is what you get out, and that using the output
again, as input into another MultipleFileProperty gives the same results.

Proper functionality in a large number of use cases is assured by vastly
improved coverage of unit tests, which was lacking previously.

Simplify parts of the MultipleFileProperty class, including the filename
generation functors I introduced that weren't doing anybody any favours.

Also kill a spurious error when LoadNexusProcessed was checking whether
it is able to load. It throws anyway so the error log is unnecessary.

This reverts commit 4fe9cd9d, and fixes
tests.

This reverts commit a9c8a59a.

Allow filenames specified with the format <<short><operator>>...<short>
where:

<short> = <dir><inst><underscore><runs><ext>
<operator> = "+" or ","
<runs> = lists and/or ranges of (possibly added) runs e.g. "12,13+14,15:20".

See header file of MultipleFileProperty for more info on syntax.

This enables workspaces loaded through the algorithm to have a history that
can be used to reproduce the workspace properly.

MWRunFiles widget of LoadDialog now remembers the exact text a user typed -
in this way any user who types "INST0-50", clicks "Run" and reopens the
dialog is not met by a huge string containing a list of all 51 fully
resolved files, but rather their original input.

Changed the vector<vector<string>> templated versions of the toString and
toValue functions to accept customised delimiters.  This enables us to add
an option for the user to turn off multifile parsing if they are crazy
enough to want to load files with a plus or a comma in their path.

Changed the inner workings of Load for multi files.  No longer calling sub
algorithms with setChild or alwaysStoreInADS.  All workspaces are
essentially created outside the ADS and managed by the algorithm, right up
until the OutputWorkspaces are set.

Tests added where appropriate, but the majority of parsing cases are already
covered by MultiFileNameParser.

WorkspaceGroup now holds shared pointers instead of workspace name. It can be used outside the ADS.