This reverts commit 5830a400.

This reverts commit e0e53de4,
since it seems that adding pkg-config as a hard dependency is
a bad idea (it's quite hard to install correctly in Windows).

(ticket #5718)

Conflicts:
	ChangeLog

Conflicts:
	ChangeLog

Keep a count of how many command line arguments we have left, so that
we know when to stop parsing.

This caused an assertion failure when we connected to a SOCKS
destination. We didn't call pending_socks_cb(), and instead went
directly to downstream_read_cb() which asserted that
'circuit_t.is_open' is true.

We now only call bufferevent_enable() on the bufferevent in
pending_socks_cb() when we finally connect to our SOCKS destination.

Also, have '--version' output the release name and not just the git
hash.

0.1.2 was skipped since I tagged without doing the release
ritual. Still getting used to this...

bufferevent_socket_connect_hostname() might fail immediately, and take
the whole circuit with it (since pending_socks_cb() and error_or_eof()
will be called). This means that we should not play with conn_t or
circuit_t after calling bufferevent_socket_connect_hostname().

Achieve that by splitting the circuit creation into three parts:
First, create the connection on the other side of the circuit. Then,
build the circuit. Finally, set up bufferevent callbacks.

Conflicts:

	src/network.c

In open_outbound{,_hostname}(), move the connection building
part to a new function called conn_connect(). Then move the
connection creation part of those functions, to
 create_other_side_of_conn().

Also, implement --help.

This adds a new one shot automake macro AX_ENABLE_HARDENING and it attempts
to add the various security hardening flags we'd like to see if they're
available. When they are not available, we simply skip adding the option.
This should ensure that we have everything we could possibly have without
adding incompatible flags or having to add a lot of weird cruft to the
configure.ac file.

Conflicts:
	ChangeLog

Fix the sed command that broke building the python test in a
build-tree, since it was specifying a path to the input that
assumed make was running in the top srcdir (re: #5310).