Fix lexc compile errors: duplicate Multichar_Symbols and mredup typo

[snomos]

While building the copy of TRmorph pulled into giellalt/lang-tur-x-ext-trmorph (as a git subtree), analyzer.cpp.lexc failed to compile with both foma's and HFST's lexc parsers.

Root cause

analyzer.lexc ends up including the multi-char symbol declaration twice:

• once directly, via its own #include "multi-char.lexc"
• once more via #include "morph.cpp.lexc" — a pre-expanded copy of morph.lexc, which itself does #include "multi-char.lexc"

Since morph.cpp.lexc is already-expanded text (no more #include directives left by the time it's pasted in), the duplication can't be prevented from within multi-char.lexc alone unless the whole thing goes through a single preprocessor pass.

Separately, morph.lexc (LEXICON Nfamily, the ALLOW_MREDUP branch) has a typo: %<N%::mredup%> has an extra, unescaped colon, giving the entry two upper:lower separators instead of one. Every other occurrence of this tag elsewhere in the file uses %<...%:mredup%> (single colon).

Fix

• multi-char.lexc: wrap the Multichar_Symbols declaration in a standard C-style include guard (the build already runs these files through cpp, so this works with the existing Makefile/makedepend.sh setup).
• analyzer.lexc: #include "morph.lexc" (raw source) instead of #include "morph.cpp.lexc" (pre-expanded), so the file is composed in one cpp pass and the include guard actually takes effect.
• morph.lexc: fix the <N:mredup> typo to match the tag used everywhere else.

Verified analyzer.cpp.lexc compiles cleanly afterwards with both the classic HFST hfst-lexc (3.17.3) and foma -f analyzer.cpp.xfst, and that the duplicate Multichar_Symbols block and the typo are both gone from the generated output.