feat: add disabled_converters parameter, converters property, and CLI flags (closes #1665)

[RhythrosaLabs]

Summary

Closes #1665

Resolves the two friction points raised in #1665 — no way to disable specific built-in converters without subclassing — with a clean, backward-compatible API addition.

---

Problem

When MarkItDown() is constructed, all built-in converters are registered unconditionally. There is currently no public surface to:

1. Security-harden a deployment by excluding dangerous converters (e.g. ZipConverter for zip-bomb protection, AudioConverter to prevent implicit Whisper API network calls on untrusted input).
2. Slim down a deployment where optional extras are intentionally not installed and the offending converter would immediately raise MissingDependencyException.
3. Inspect which converters are active without reaching into the private _converters list.

---

Changes

packages/markitdown/src/markitdown/_markitdown.py

What	Details
MarkItDown.__init__(disabled_converters=…)	New keyword-only parameter. Accepts any iterable of DocumentConverter subclasses. Each listed type is skipped during enable_builtins(). Validation is eager: raises TypeError if any element is not a DocumentConverter subclass.
MarkItDown.disabled_converters property	Returns the frozenset of types that were excluded at construction time.
MarkItDown.converters property	Returns a read-only tuple[ConverterRegistration, …] snapshot sorted by priority — no more reaching into _converters.

packages/markitdown/src/markitdown/__main__.py

What	Details
--disable-converter CONVERTER	Repeatable flag; resolves class names against markitdown.converters. Unknown names exit non-zero with a clear message.
--list-converters	Prints all built-in converter class names and exits 0.

packages/markitdown/src/markitdown/__init__.py

ConverterRegistration is now exported from the top-level package so callers can type-annotate against it.

packages/markitdown/tests/test_disabled_converters.py (new file)

22 tests covering:

• Default construction registers all built-ins
• Single / multiple / empty / frozenset disabling
• TypeError on non-subclass, instance, or None
• disabled_converters and converters properties
• Snapshot immutability of converters
• Behavioural: disabled converters are not invoked
• enable_builtins=False + manual enable_builtins() pattern
• CLI --list-converters, --disable-converter, unknown name detection, multi-flag

---

Usage examples

Python API

from markitdown import MarkItDown
from markitdown.converters import ZipConverter, AudioConverter

# Security hardening: never process zips or audio on this endpoint
md = MarkItDown(disabled_converters=[ZipConverter, AudioConverter])

# Inspect what is active
for reg in md.converters:
    print(type(reg.converter).__name__, reg.priority)

# What was excluded?
print(md.disabled_converters)
# frozenset({<class 'ZipConverter'>, <class 'AudioConverter'>})

CLI

# See all built-in converter names
markitdown --list-converters

# Disable ZipConverter and AudioConverter for this conversion
markitdown --disable-converter ZipConverter --disable-converter AudioConverter document.pdf

---

Before / After

Scenario	Before	After
Disable ZipConverter	Subclass MarkItDown, override enable_builtins()	MarkItDown(disabled_converters=[ZipConverter])
Know what is registered	md._converters (private)	md.converters (public, sorted tuple)
Know what was excluded	Not possible	md.disabled_converters (frozenset)
CLI disable	Not possible	--disable-converter ZipConverter
CLI list names	Not possible	--list-converters

---

Checklist

• Backward-compatible — disabled_converters defaults to None / empty frozenset; all existing code works unchanged
• enable_builtins=False + later enable_builtins() pattern also respects the disabled set
• Eager input validation with a descriptive TypeError
• 22 new tests; all existing tests pass (pre-existing test_speech_transcription failure is unrelated, requires local Whisper setup)
• No changes to converter internals or converter-selection logic