Add -config option to ocamlrun

[dra27]

The main purpose of this PR is to add a -config option to ocamlrun similar to the one already in ocamlc/ocamlopt. This is loosely related to the work I'm doing with relocating the compiler, but independent enough to push it now - it's obviously non-urgent. ocamlrun -config gives output like:

version: 4.11.0+dev0-2019-10-18
standard_library_default: /usr/local/lib/ocaml
standard_library: runtime
int_size: 64
word_size: 63
os_type: Unix
host: x86_64-pc-linux-gnu
flat_float_array: true
supports_afl: true
windows_unicode: false
supports_shared_libraries: true
no_naked_pointers: false
profinfo: false
profinfo_width: 0
exec_magic_number: Caml1999X027
shared_libs_path:
  /usr/local/lib/ocaml/stublibs
  /usr/local/lib/ocaml

Options are exactly the same as ocamlc -config, with four exceptions:

• supports_afl indicates that the runtime has the shared memory support for execution with AFL, which is independent of using a compiler which had AFL instrumentation turned on for a particular build. cc @stedolan to ensure I got the detail of that correct
• shared_libs_path which contains the contents of caml_shared_libs_path after parsing ld.conf and any -I parameters on the command line (this is the feature I particularly wanted).
• no_naked_pointers and profinfo/profinfo_width

There's also a little shopping list of other things related to ocamlrun's argument parsing:

• Support -- so you can now have bytecode programs which begin with - if you, um, want
• Support -Ifoo as -I foo (previously, this was interpreted as -I
• Update some missing things in the manual
• Use puts instead of printf("%s\n", …) in a few places in startup_byt.c
• Strictly recognise command line parameters (previously, the first letter was checked only, so -backtrace was equivalent to -b - -backtrace is now an error).
• Display a slightly improved message for ocamlrun -I (indicate that -I expects a <dir> rather than that no bytecode file was specified).
• Add -help/--help

[stedolan]

The AFL bit looks good to me!

[dra27]

I've broken this PR up into parts in a bid to get it merged. This PR also includes #10098 and #10101 which need merging first.

Note that this PR works without #10103 (since if WINDOWS_UNICODE is not defined, it will default to 0 according to the specs for the C preprocessor) but I think it may fail one of the sanitizer checks.

[xavierleroy]

More nitpicking on the parts that have not been sliced off.

[xavierleroy]

Looks good to me, once all the various PRs are put back together.

[dra27]

Rebased, squashed, extra bits removed and ready for a CI-approving merge, thanks!