CLI: add per-pass removal toggles (--no-content-patterns etc.)

[kaznak]

Problem

Defuddle exposes a set of remove* options in its JS API (DefuddleOptions in src/types.ts):

• removeExactSelectors (default true)
• removePartialSelectors (default true)
• removeHiddenElements (default true)
• removeLowScoring (default true)
• removeSmallImages (default true)
• removeContentPatterns (default true)
• removeImages (default false)

The CLI never wires any of them up. src/cli.ts builds defuddleOpts with only { debug, markdown, separateMarkdown, language }, so the CLI runs with the full pipeline locked on and there is no way to disable a single step without forking.

For users developing site-specific extractors this is the recurring question: which removal pass dropped this element? The cleanest debugging workflow is "run with that pass off, diff against the default output." Without per-pass toggles, the only way to answer that is to fork defuddle and add console.log calls.

Proposal

Add seven CLI flags on parse, each binding 1:1 to an existing remove* API option. Six are --no-* form (matching the API default of true); one is positive (removeImages defaults to false):

--no-content-patterns    Keep boilerplate patterns (read time, breadcrumb, metadata lists, newsletter signups, …)
--no-low-scoring         Keep low-scoring elements
--no-exact-selectors     Keep EXACT_SELECTORS matches (ads, scripts, JW Player, …)
--no-partial-selectors   Keep PARTIAL_SELECTORS matches (class/id containing "ad-", "sidebar", "comment", …)
--no-hidden-elements     Keep CSS-hidden elements (display:none, visibility:hidden, opacity:0)
--no-small-images        Keep small images
--remove-images          Strip <img>/<picture>/<figure>

Example workflow for extractor development:

# baseline: see everything (use sister --debug PR to log what would be removed)
defuddle parse 'https://example.com/article' \
    --markdown \
    --no-content-patterns \
    --no-low-scoring > baseline.md

# default pipeline output
defuddle parse 'https://example.com/article' \
    --markdown > pipeline.md

# diff tells you what the pipeline stripped
diff baseline.md pipeline.md

Implementation note

Defuddle merges options as:

const options = {
    removeExactSelectors: true,
    /* ...other defaults... */
    ...this.options,
    ...overrideOptions
};

So an explicit undefined from this.options shadows the default true and silently disables the removal pass. The CLI therefore only inserts a remove* key when the user actually set the flag — done via a small maybe() helper that emits an empty object for undefined values, instead of { key: undefined }. This keeps both call paths working:

• via commander (--no-content-patterns absent → options.contentPatterns === true → removeContentPatterns: true set, but explicitly matching the default — no-op)
• via direct API import (parseSource('-', {}) → options.contentPatterns === undefined → key omitted → default true applies)

Tests

tests/cli.test.ts adds:

• Option introspection: all seven flags are registered on parse.
• Behavior: --no-content-patterns keeps a 2-item trailing <ul> that the metadata-list heuristic would otherwise strip.
• Behavior: --remove-images strips an inline <img> that the default pipeline keeps.

All 371 existing tests continue to pass.

Notes

• Combines naturally with --debug (sister PR feat/cli-debug-output): turn a single pass off and use the debug log to see what the rest of the pipeline still removes.
• Combines naturally with --extractor (CLI: add --extractor flag and defuddle/extractor public entry #328): inspect what a custom extractor's output looks like before any removal pass touches it (--no-content-patterns --no-low-scoring --no-exact-selectors --no-partial-selectors).
• The flags map 1:1 to API options; no new behavior is added to the library.