feat(document): show text and code documents in a code viewer

[pirhoo]

Text, code, JSON, XML and HTML documents now open in a read-only code viewer with syntax highlighting, line numbers and search.

• feat: add a read-only code viewer with language detection
• feat: search inside the code viewer, including very large files
• feat: route text, code, JSON and HTML documents to the new viewer
• fix: stop downloading sources larger than 50 MB, even when their size is unknown
• fix: decode sources with the encoding found at indexing
• fix: highlight matches correctly after emoji and for Greek words
• fix: keep blurred documents out of search
• refactor: remove the JSON tree viewer

Preview

[caro3801]

Very nice improvement of the code display. I only have suggestions for this PR considering that the cases mentions have low probability of occurring, let me know what you think.

[caro3801]

suggestion: the raw content type can contain multiple info
(application/xml; charset=utf-8) the exact match branch would then and lands on "Preview is not available", while text/...; charset=... sails through on the prefix test. This index demonstrably stores parameters (isTweet at line 412 matches
'application/json; twint').
Could we strip the parameter once before testing? codeLanguage.js:14 already does the same split(';') , so the shape is familiar (may be factorized).

Suggested change

	return this.contentType.indexOf('text/') === 0
	|| this.contentType.endsWith('+xml')
	|| this.isJson
	|| codeTypes.includes(this.contentType)
	}
	const mimeType = this.contentType.split(';')[0].trim()
	return mimeType.indexOf('text/') === 0
	|| mimeType.endsWith('+xml')
	|| this.isJson
	|| codeTypes.includes(mimeType)

[pirhoo]

Done, the type is now stripped of its parameters before the checks.

[caro3801]

suggestion: foldWithSourceIndexes builds two offset arrays over the term that are destructured away immediately.
foldForFilter (added in this diff, already imported on line 1) returns the same string without them. Worth dropping foldWithSourceIndexes from the line 1 import once this changes, since it becomes unused in this file.

Suggested change

	const { folded } = foldWithSourceIndexes(term)
	const folded = foldForFilter(term)

[pirhoo]

Done, and the unused import is removed.

[caro3801]

question (perf): any particular reason to drop the push loop ? These two lines allocate two throwaway arrays plus two spreads per character, on the path that folds for every candidate chunk. A plain push loop keeps the behaviour exactly and drops the allocations.

Suggested change

	sourceIndexes.push(...Array(units).fill(index))
	sourceEnds.push(...Array(units).fill(end))
	for (let unit = 0; unit < units; unit++) {
	sourceIndexes.push(index)
	sourceEnds.push(end)
	}

[pirhoo]

No reason, the spread version was just extra allocations. The loop is back.

[caro3801]

suggestion: The superseded-load guards around load are careful and correct on every path, including abort, 404, too-large and unmount. The blurred-content watcher is the one async path that did not get the same superseded-load treatment, so a fast document switch can strand the new document behind the previous project's banner with search disabled.

Could we re-check the document identity after the awaits, the way isCurrentLoad does for the source?

Suggested change

	watch(toRef(props, 'document'), async (document) => {
	blurred.value = await isBlurred(document)
	blurredContent.value = blurred.value ? await getBlurredContentBanner(document) : null
	}, { immediate: true })
	watch(toRef(props, 'document'), async (document) => {
	const value = await isBlurred(document)
	const banner = value ? await getBlurredContentBanner(document) : null
	// A document swapped in while this one resolved owns the banner now.
	if (document !== props.document) {
	return
	}
	blurred.value = value
	blurredContent.value = banner
	}, { immediate: true })

[pirhoo]

Done, a superseded result is now dropped. There's a new test for a fast document switch.