Ignore invisible whitespace in text matching

[passsy]

Fixes #138

spotText and the AnyText text selectors now match what a user sees: invisible characters and characters that render as a space are ignored, so tests can use plain characters instead of escapes.

// Given Text('foo\u{200B}bar')  → ZWSP inserted for line breaks
spotText('foobar').existsOnce();

// Given Text('12\u{202F}345\u{00A0}€')  → narrow + non-breaking spaces
spotText('12 345 €').existsOnce();

Normalization rule

normalizeVisibleText(String) applies a clear, two-part rule:

• Removed (invisible, no meaning): Zero Width Space U+200B, Soft Hyphen U+00AD, Word Joiner U+2060, BOM U+FEFF
• Folded to a regular space: every Unicode space separator (Zs) — Non-Breaking Space U+00A0, Narrow No-Break Space U+202F, thin/em/figure spaces, ideographic space, …

Deliberately left untouched, because they carry meaning or structure: Zero Width Joiner / Non-Joiner (emoji sequences, Indic/Arabic shaping), bidi controls, the U+FFFC WidgetSpan placeholder, tabs and line breaks.

Matching the exact characters

When a test needs the literal bytes, AnyText now exposes a rawText property alongside the normalized text, generating a raw selector family:

// Text('foo\u{00A0}bar')
spotText('foo').withText('foo bar').existsOnce();           // normalized
spotText('foo').withRawText('foo\u{00A0}bar').existsOnce(); // exact

New: whereRawText, withRawText, hasRawText, getRawText, hasRawTextWhere on WidgetSelector<AnyText>, plus public normalizeVisibleText, extractTextContent(Element), the TextContent holder (raw/normalized), and zeroWidthSpace/nonBreakingSpace constants.

Scope

The concrete spot<Text>() / spot<SelectableText>() / spot<EditableText>() selectors read their native diagnostic props and are unchanged. Use spotText / spot<AnyText> for the normalized behavior.

Builds on the idea from #139 (thx @MichaelTamm), extended to a complete character set with a raw escape hatch.