Split dwarf.c into per-section files under format/dwarf ##bin

[phix33]

• Mark this if you consider it ready to merge
• I've added tests (optional)
• I wrote some lines in the book (optional)

Description

libr/bin/dwarf.c had grown to 5043 lines holding eleven unrelated parsers. This moves it to libr/bin/format/dwarf/ and splits it by section:

$ wc -l libr/bin/format/dwarf/*.c
   94 abbrev.c     .debug_abbrev
  111 aranges.c    .debug_aranges
  543 dump.c       the id dumps and their name tables
  235 dwarf.c      section lookup and the shared byte readers
  561 eh_frame.c   Itanium LSDA and .eh_frame
  427 index.c      the DWARF 5 str_offsets/addr/loclists_base resolver
 1239 info.c       .debug_info DIEs and compilation units
 1197 line.c       .debug_line
  269 loc.c        .debug_loc
  329 loclists.c   .debug_loclists

• Every function body is moved verbatim; git diff --color-moved=dimmed-zebra shows only moved blocks plus includes, a private dwarf.h with the READ* macros, the section enum, two typedefs and sixteen R_IPI dwarf_* prototypes, and the static drops those need. The five generically named helpers (get_section, ...) get the dwarf_ prefix first so nothing generic leaves the file; the lib builds with -fvisibility=hidden, so nm -D of libr_bin.so is unchanged.
• Two small commits land before the move so the split itself is a pure move: abbrev_find replaces three identical bsearch pairs in the CU loops, and get_section_bytes only reads an Elf_Chdr when the object is an ELF, instead of casting whatever bin_obj is to ELFOBJ for any zdebug/SHF_COMPRESSED section (non-ELF compressed sections now return NULL, as the header check already made them).
• format/dwarf/deps.mk follows format/pdb/deps.mk; meson lists the ten files.
• The eh_frame/LSDA code keeps its r_bin_dwarf_ names and i/private.h declarations.

id, CL, idx, tl, t and afv after aaa are byte-identical to master on ten DWARF fixtures (2.4 M lines compared), and the load cost is unchanged. The new db/formats/dwarf case pins the SHF_COMPRESSED path on the .debug fixture debuglink already uses (123228 rows); it fails when the decoder is skipped.