svn:r1808

svn:r1801

svn:r1800

bugfix: add a circ->resolving_streams field, and make dns resolving                            streams not actually in the connection_array until they're done resolving.

bugfix: actually complain if we duplicate mark-for-close a circuit

add more logging for relay ends that claim dns resolve failed, so we can
find out why they're not being retried.


svn:r1798

svn:r1797

svn:r1796

svn:r1795

svn:r1794

svn:r1793

svn:r1792

unused portions of relay payloads are nul-padded, not filled with
random bytes.


svn:r1791

svn:r1790

svn:r1789

Only connection_add connections once they have conn->s sett; refactor code around this.  Should make stuff more bulletproof.


svn:r1788

svn:r1787

otherwise they might have a socket -1, and if we leave them around
they'll muck up poll/select.

i think this was the cause of our win32 and os x fakepoll crashes,
and probably would cause other errors down the road.


svn:r1786

svn:r1785

svn:r1784

svn:r1783

svn:r1782

svn:r1780

svn:r1778

svn:r1777

svn:r1776

svn:r1775

svn:r1774

svn:r1771

svn:r1770

svn:r1769

svn:r1768

svn:r1767

svn:r1766

mark the circ for close if circuit_extend fails because the next
hop is not up yet; don't just return -1


svn:r1765

svn:r1764

svn:r1763

svn:r1762

svn:r1761

svn:r1760

svn:r1759

svn:r1758