Construct makecontext() arugments properly on systems
like arm7l that have 32 bit pointers.

Use %zu format specifier to refer to a size_t argument
to avoid compilation errors on arm7l.

Use %PRIi64 not %ld to refer to int64_t argument to avoid
compilation errors on arm7l.

Problem: running 10,000 coprocs causes coproc_create()
uses 10000*2mb = 20gb of VM, resulting in ENOMEM on
a system with 32-bit pointers.

The requested amount of memory cannot be represented in
32-bit pointers.  In addition, the default vm.overcommit_memory
setting (0: heuristic overcommit handling) may reject an
allocation greatly in excess of physical memory.
Run 500 instead of 10K coprocesses, which is still
a reasonable test.

In addition, clean up the test so that it doesn't produce
a series of cascading segfaults and obscure failures when
this happens.  Fail the first test and continue on with
however many coprocs were able to be allocated.

Problem: broker attr test fails on arm7l architecture.

Change strtol() to strtoul() when converting UINT_MAX - 1.

Backport of grondo/lua-affinity#2 to copy of lua-affinity in
flux-core.

Fixes flux-framework#975

Problem: flux-ping segfaults in flux_rpcf_multi() on arm7l,
which has 32-bit struct timeval members.

The "I" format specifier refers to a 64-bit argument in
json_vpack_ex() and passing 32 bit arguments triggers a segfault.
The segfault is avoided in flux-ping by casting tv_sec and tv_usec
arguments to uint64_t.

Problem: 5s timeouts are too tight on tests 1 and 6 on
are too small for some systems.

Actual runtimes on raspberry pi were 5.7s and 8.4s,
respectively.  Increase timeouts to 10s and 20s.

Problem: wlog_fatal() sends log messages to flux_log() or
vfprintf (stderr, ...) depending on context.  The former
takes messages without newlines; the latter requires a newline.

Append a newline character to the stderr stream when that
mode is selected.  Drop trailing newlines from a couple
of wlog_fatal() calls to be consistent with the majority
of other calls.

Problem: zeromq-4.2.1 reports EHOSTUNREACH as "Host unreachable",
but "No route to host" is canonical on Linux and we have some
tests that depend on it, so remap here.

Problem: zeromq-4.2.1 reports EHOSTUNREACH as "Host unreachable",
but "No route to host" is canonical on Linux and we have some
tests that depend on it, so remap here.

Problem: on arm7l architecture, wrexecd behaves erratically.

Add printf-style format attributes to internal varargs debugging
functions to enable type checking, then fix a couple places where
the format specifier (%lu) implied 32-bit on this architecture,
but the argument was 64 bit.  Also fix non-const format specifier
warnings.

Due to a typo in which variables were interchanged, an error
handling code path in cache_store_continuation() could trigger
a SEGFAULT on arm7l test system.

Fix the typo.