gdb: Use multiple minimal_symbol_readers in mipscoff_symfile_read

gdb: Use multiple minimal_symbol_readers in mipscoff_symfile_read

Currently, mipscoff_symfile_read uses a single minimal_symbol_reader to
get all minimal symbols from mdebug-formatted debuginfo, and from
alphacoff. This pattern has been around since minimal_symbol_reader has
been introduced, and from own research, there's no need to use the same
reader. This made it so mipscoff_symfile_read could call
mdebug_build_psymtabs directly, since the latter needs a reference to
the minsym reader object. The issue is that future commits need a
unified entrance point to read debuginfo, and this pattern is very
different to how elf does mdebug reading.

In fact, the elf mdebug reader does some preparatory steps and then
calls mdebug_build_psymtabs, so what the mips version does is just
spread these preparatory steps through the mipscoff function instead.

To make it easier for future commits to query debuginfo support
dynamically (as opposed to assuming things at compile time), this commit
introduces a new mipsmdebug_build_psymtabs function, which does similar
preparatory steps as elfmdebug_build_psymtabs. It is added to
mdebugread.c to help with maintaining a separation between reading an
objfile (in mipsread.c) and its debuginfo (mdebug), so that in the
future we have an easier time selectively disabling debuginfo formats
at compilation time. This should have no visible changes for the end
user.

The new function must receive the pointers to ecoff_debug_swap and
ecoff_debug_info because finding those structures based on the bfd
object necessitates including the headers libcoff.h and libecoff.h,
and those headers can't be included at the same time as libaout.h
- which mdebugread.c already uses.

Approved-By: Tom Tromey <tom@tromey.com>