As make test creates a cellbuild, the tests should look for modules in the
cell directory.

This removes the need for activating an environment as part of any custom
test targets.

Apparently make 3.81 treats the following
```
define foo =
bar
baz
endef
```
differently than make 4.2.1 (most notably, the equals sign).

The original e3 build installed the module.Makefile in the module
source directory, and then runs (essentially)
```
$ make -C module -f module.Makefile ARGS
```
However, there is no real need for it to copy it there instead of
simply using that makefile from the parent directory. This simplifies
a bit of the junk that gets added around during the build process.

The error messages

    error: could not lock config file .git/modules/foo/config: No such file or directory
    warning: Could not unset core.worktree setting in submodule \'foo\'

can be seen for all non-local mode wrappers (source code as git
submodule).

This happens if doing a `git submodule deinit` before a submodule has
been initialised. If a submodule has been initialised, but then is
deinitialised twice, the second error message shown above is still
displayed.

This fixes the above cases by instead doing a forced recursive deletion
of the submodule directory.

There is sufficient added complexity to manage this, for limited value.
If there are issues with tests failing that require debug symbols, one
can manually modify the pytest-generated wrappers to add the symbols
back.

Similar to the change to driver.makefile to handle multiple architectures,
we use make's default behaviour to fail a task if any dependency fails.

This is not really necessary; you can always re-initialise the submodule
or check it out again.

These were intended originally to allow a hybrid build system: you
could build an e3 module as the dynamically loaded shared library,
but you could also run `make epics` and build it as a vanilla EPICS
module.

However, this has never really worked, nor has it been supported. It
may in the future be supplanted by properly using the EPICS build
rules in lieu of the use of driver.makefile, but that is for a future
change.

The issue was that `git status --porcelain` formats its output in such
a way that `make` no longer passes the variables correctly. The solution
is simply to no longer pass the `git status` data.

Note that the git status data already was not that sufficient in terms
of recognising what changes had been made, since all it provided was
information as to which files had been modified, not how they had been
modified.

This removal allows us to remove the `hdrs` target from every module; there
were only two modules that still used it, and they have both been fixed.

The purpose of this is to remove the dependency on tclx, a non-standard
package.

If a user runs the build from the E3_SITEMODS_PATH, then it can happen that
false dependencies get registered due to how the generated .d files are
parsed to determine module dependencies. Specifically, if you have header
file being used that is located at

${E3_SITEMODS_PATH}/e3-module/module/src/header.h

and it is _not_ being installed, then the dependency parsing script will
see a dependency as an absolute path, and so register that your module
depends on the module `e3-module` version `module`, which is obviously
not what was intended.

While we don't want to stricly _prevent_ a user from building there, it
seems wise to _warn_ a user that there may be unintended consequences.

The test rule for e3 modules actually does a cellinstall and run the
tests inside this temporary installation. The cellinstall needs to
overwrite some path variables to work this variables were not accessible
through the test rule as they are just changed if cell is on
MAKECMDGOALS. This made impossible to use the standard path variables
inside module's custom test rules.

To fix it the same mechanism used in cellinstall is implemented for test
and the variables paths are now overwritten. This makes possible ie. to
use E3_MODULES_INSTALL_LOCATION_BIN inside custom test rules.

This commit also introduces an unwanted behavior. As the test rule is
now overwriting the standard path variables it cannot longer be along
with the install rule. For example, running `make test install` will
install the module inside the test dir. This is a known issue for
cellinstall as well and a restructure of the Makefile code should be
made to isolate the rules.