The venv module is part of standard python, so let's use it instead of
virtualenv, which requires its own installation step.

Make is not a great taskrunner.  There are some tasks that aren't
really about building and we don't need make for it.  So here's a simple
task runner where we can put those things.

To start, we have a 'watch' task, which just watches the doc for changes
and rebuilds as needed.  We also have a task that checks the links in a
pdf for 200 responses and complains if it doesn't get one.

Also, add a Makefile target to establish bin/dosh in new doc dirs.  That
target isn't automatically hit.  The user has to do it manually.  Maybe
we'll tie it in so it just gets added (like venv), but for now it's not
a crucial part of the workflow, so we don't do it.

The bin/dosh we stick in dirs is not just a copy and not just a symlink.
It's a wrapper around our dosh.  This lets the user inherit our targets
but also add doc-specific changes and new tasks.

Before now, building "foo.pdf" would instead build "foo-rREVNUM.pdf"
(where REVNUM is determined by 'get_revision') and symlink "foo.pdf"
to that.  Although this had its uses, it also would occasionally cause
problems, primarily accidentally copying the symlink to somewhere else
and ending up with a broken symlink at the destination.  After
discussing it yesterday, James and I decided to switch to the simpler
"just build foo.pdf" method, and that's what this commit does.

At least right now there is a moment in Debian 'testing' in which
there is no 'python' command, only 'python3'.  Since we are already
explicitly finding 'python3' for PYBIN anyway, just use that.

I don't actually use the venv, and I was tired of it littering.
Virtual environments are probably best left to the user to configure,
and not for us to force on the user.  It's still nice to have it as a
backup, but if the user has a way to satisfy those dependencies
without our intervention, don't intervene.

There are a few ways to supply the python submodules that aren't in a
local venv.  One might install systemwide, use direnv, or use a
centralized venv for all of ots, etc.  I use direnv, but this change
makes room for any scheme a user might want.  We force our own venv
only as a last resort.

This fixes the problems described in issue #11, but is not a solution
that anyone should invite home to dinner.  See issue #15 for details.

This follows up to commit 018499ec, which introduced a pytest import.

This might be useful for diffing against when your doc suddenly
doesn't build and you don't know why.  This doesn't save all the files
that go into making the doc.  It would be interesting to try to
snapshot them in some way.  It might even be interesting to collect
all the various files from all over and stik them in a dir so that
every build also makes a minimal one-dir version for slinging around.
Alas, this does none of that.

 * Make -rXXXX versions for all builds, not just drafts
 * Add draft plugin
 * In template, set draft to trigger on both True and 'True'
 * Move path into PIPELINE definition
 * Silence some echo
 * Use .tex instead of intermediate_ltx
 * Simplify some bash
 * Let the plugin handle draft instead of passing it as an option
 * Cleanup .tex files

We can now do redacted and draft at the same time.

Refactor the jinja rendering and the redaction as a pipeline of
filters that run as needed.

It works, but needs some cleanup and documentation

When a venv is created, 'click' will now be installed into it.

This also removes the "dd" that was accidentally introduced into the
variable's value in commit 8463ab1fd21e, the same commit that made the
variable unused (I think we can guess the vi-ism that happened here).
So if we're ever restoring this variable, remember to remove the typo.

Those are backup files or autosave files, and they muck up the build
process, especially if they're a link to a file that doesn't exist.

Don't print all the possible 'make foo.pdf' commands when there are so
many source files (not all of which are necessarily even buildable on
their own) that the list would be very long.

Move the LaTeX template files and the get_revision script from OTS's
private SVN tree to here.  Rewrite README.md.  Clean up the Makefile,
moving all documentation into README.md removing the "VIEW=true"
feature as discussed with James.

Use spaces instead of TABs for the left-side whitespace on line inside
the shell expression.  Break the first line into two lines, and on all
lines move the continuation backslashes leftward to stay within 80 cols.