Add load_area and a JSON CLI; fix section skipping and record diagnostics

[ctoth]

Summary

• cli.load_area(path, dialect=None) (fixes the red CI: test_medievia.py calls it). dialect is a DIALECTS key (rom, merc, smaug, circle, tba, medievia, godwars, swr, coffeemud), an area-file class, or None. An explicit dialect is parsed as-is and its errors propagate. With None, the detected dialect runs first, then ROM, Merc and SMAUG (files only). The first parse with rooms wins. If no parse has rooms, the first parse that succeeded is returned. If nothing parses, the detected dialect's error is raised. If detection itself failed, its ValueError is raised, chained from the fallback error that got furthest into the file.
• CLI: area-reader [--type auto|rom|merc|smaug|circle|tba|medievia|godwars|swr|coffeemud] <path>. On success it prints JSON to stdout and exits 0. On failure it prints one JSON line to stderr and exits 1: {"error", "file", "line", "column", "section"}. python -m area_reader.cli works, and logging goes to stderr.
• ParseError now has reason, filename, and 1-based line/column fields, plus section for the text readers. The formatted message is unchanged.
• Unknown sections (skip_section truncates unrecognized sections at the first '#' in their body (silent data loss / misparse) #5): a section is skipped up to the next line that starts with # plus an uppercase name (or #$), so a # inside the body no longer cuts it short.
• Diagnostics (Repeating vnums between rooms/npcs/objects #2, partial): a duplicate vnum in rooms/mobs/objects still overwrites the earlier record, but each overwrite is now recorded as {"kind": "duplicate_vnum", "family", "vnum"}. Each skipped section is recorded as {"kind": "skipped_section", "section"}. Every reader serializes the list as "diagnostics" in as_dict/as_json. skipped_sections stays as it was. Duplicate detection covers the text readers (ROM/Merc/SMAUG/SWR/GodWars); the Circle-family and CoffeeMud readers emit an empty list.
• Dead code (Dead loader code: SmaugAreaFile.load_room calls nonexistent read_room_data; jump_to_section mispositions on missing section #10): removed SmaugAreaFile.load_room (it called a nonexistent read_room_data) and its read_line. jump_to_section now matches the header at a line start and raises ParseError when the section is missing. Merc and SWR share model.native_reset_arg2_suffix. The Medievia suffix behaves differently, so it is kept.

detect_area_type is unchanged.

Fixes #5
Fixes #10
Partially addresses #2 (records duplicates; does not change overwrite behaviour)

Verification

• Full suite: 1077 passed
• Corpus sweep (321 files): 221 ok / 100 fail, the same as master. load_area returns the same parser as the sweep's fallback loop for every file (0 mismatches).

🤖 Generated with Claude Code

https://claude.ai/code/session_01LsTyuYeZHogKAnGhufs9gN