Notably, pass around the location of the xcvb-program we create in the cache.

* Use define-command to define and register a command
* Introduce the notion of workspace to store intermediate and target files.
* Update Makefiles for new intended output paths

as I was tired of completion conflicts with driver.lisp.

Move fake-release-directory to Lisp.
Fix enough-namestring in Makefile.

driver: fix back the allegro run-program after latest changes
tests: refactoring so xcvb-driver-test can be run independently
other: fix some missing renamings of escape-shell-token.

Converted more ERROR calls to USER-ERROR and fixed up the message to
be more clear.

When the simple-dispatcher constructs a dispatch set of forms, the
name of the dispatcher is now embedded into it. This makes it
much easier to find _which_ simple dispatcher interpreter might be
missing how to evaluate a form.

* extract fasl type from target:
 at least lispworks depends on pathname type and its fasl type varies!
* rucksack not available on clisp (and needed fixes; see our xcvb repo of it)
* accept compile-file failure semantics of clisp
* suppress duplicate entries for a directory in the source-registry:
 The first one wins, with its recurse and exclude flags.
* fix hello and test/runme
* fix various option parsing mistakes

	settings known to xcvb or the individual value of a single one.
Change the default object-directory location to be
	$HOME/.cache/xcvb/common-lisp/<lisp-implementation-version>/
Added format! in driver.lisp to help with debugging and forcing output.
Debugged CMUCL support in driver.
Fixed implementation-identifier support in extract-target-properties.
Allowed Makefile in example-1 to parameterize on lisp implementation.
Refactored how option-specs are specified for command line options.
Removed dead code paths from simple-build backend that dealt with
	"xcvb.mk" and the makefile output file.
Muffled warnings when loading asdf.
Added target-asdf-version setting which you can query with show-settings.
Started DONE.rest

since ASDF was running the old uninterned perform method while ironclad was
adding its :around method to the new ASDF perform symbol and generic function
that is not active yet.

Try to be less dumb when deciding whether to use enough-namestring.

TODO: need to better control the cwd vs *default-pathname-defaults*

Converted the makefile style transformation of a computation into
a command into a more generalized external command form of the
computations.  Condensed functionality from driver-commands.lisp and
makefile-backend.lisp into external-commands.lisp. This is because at
this time both the makefile backend and a new run-program back end will
use them.

Added a run-program based backend called "simple-build" which will execute
the commands necessary to build the dependencies in the graph. At this
time it is a serial execution of the dependencies. The functionality is
not yet complete and currently broken as of this checkin.

Fixed up the xcvb.asd file to denote the new dependencies.

Notably fixed a typo in removing feature*s*, as noticed by Theam Yong Chew. Thanks!

Also, supersede some sbcl contrib asdf systems with appropriate requires.

refactor vp handling, using a miniDSL default-vp-for-fullname.

Using a new lisp-object-grain class for .o, and fasl-grain for .fas.
Idea: for .515 I will undo much of that and use CFASL for .fas and FASL for .o

TODO for ECL: distinguish linkable object files from loadable shared objects.

to avoid staging conflict with the variable used to start XCVB itself,
when compiled with CCL. Similarly with $XSCBL, etc. Sigh.

* fix various bugs left over from .507.
* back out generation of Lisp files in object tree
 until we can provide build processes with a mapping from name to pathname.
* pass self-test.

* after profiling, too much time was spent in checking directories to make.
 Using a hash-table of implicitly (or explicitly) made directories
 as well as a list of explicitly made directories.
* use asdf-dependency-grovel::normalized-component-name
 for names of components included.
* Idea: give a number to each grain, then use FMIM for mergeable sets of
 grains (if lots of reuse - use hash-consing of sets of grains?).