refactor: generate invisible char regex programmatically from data structures

[Fieldnote-Echo]

Problem

The invisible character regex in _invisible.py is manually concatenated from 9 separate set/range sources (~20 lines of string concatenation). This has two risks:

1. Silent regression — A typo, merge conflict, or range edit can shrink coverage with no test failure (until the regression test from refactor: generate invisible char regex programmatically from data structures #8 catches it)
2. C0 range fragility — The three C0 sub-ranges use hardcoded indexes (C0_CONTROL_RANGES[0], [1], [2]). Adding or removing a range silently breaks the regex.
3. Set iteration order — Fixed in 0.2.0 via sorted(), but the root cause is that regex construction shouldn't depend on container iteration order at all.

Proposed Solution

Replace manual string concatenation with programmatic generation:

# Single source of truth: all invisible codepoints as a flat sorted list
_ALL_INVISIBLE_CODEPOINTS = sorted(
    ZERO_WIDTH_CHARS | FORMAT_CHARS | MONGOLIAN_FVS_CHARS | BIDI_CONTROL_CHARS
    | {chr(cp) for start, end in [VARIATION_SELECTOR_RANGE, TAG_BLOCK_RANGE, ...] 
       for cp in range(start, end + 1)}
    | {chr(cp) for start, end in C0_CONTROL_RANGES for cp in range(start, end + 1)}
    | {chr(cp) for cp in range(*C1_CONTROL_RANGE)}
)

# Generate regex from the flat set — collapse contiguous runs into ranges
INVISIBLE_RE = _build_char_class(_ALL_INVISIBLE_CODEPOINTS)

This makes the regex self-documenting, auditable, and immune to ordering/indexing bugs.

Context

Surfaced by Copilot and Grippy code review on PR #7. See also the regression test added in 0.2.0 that asserts all 492 codepoints are matched.

Scope

• Refactor _invisible.py regex construction
• Keep all existing character sets as named constants (documentation value)
• Verify regex matches identical codepoint set before and after
• Target: 0.3.0

[OnlyTerp]

I'd like to work on this issue. I've analyzed the codebase and have a fix ready.

[Fieldnote-Echo]

@roberthallers Welcome, we are happy to have you - go ahead and open the PR and I’ll review it. Thank you for the interest.