[iliev-blnk]

Fixes #142, fixes #143

Cause

processIndicRearragement runs the verb of every entry on whatever range firstGlyph..lastGlyph holds at that moment:

• MORX-3 ([morx] Rearrangement without marked sequence #142): verbs run before any glyph is marked. firstGlyph and lastGlyph are both null, so swap() splices at index 0 and moves glyphs from the start of the text.
• MORX-5 ([morx] Rearrangement underflow #143): the marked range is shorter than the glyphs the verb moves, e.g. ABxCD => CDxAB on a range of 1 to 3 glyphs. swap() then reaches past the range and moves R, Y or Z too.

Fix

This follows HarfBuzz, which matches CoreText here:

• A verb runs only when a range is marked and it holds every glyph the verb moves (1 for Ax => xA, up to 4 for ABxCD => DCxBA). Without MarkFirst, the range starts at the first glyph.
• MarkLast at the end of the text marks the last glyph, as HarfBuzz clamps end to the buffer length.

	master	this PR	HarfBuzz 15.0.0
MORX-3 ABXCD4	X C D A B four	A B X C D four	A B X C D four
MORX-5 PQRAXYZ⓬	P Q Y Z X R A one_two	P Q R A X Y Z one_two	P Q R A X Y Z one_two
MORX-5 PQRABCXYZ⓯	P Q R C X A B Y Z one_five	P Q R A B C X Y Z one_five	P Q R A B C X Y Z one_five

Across all 174 MORT/MORX cases of text-rendering-tests, this PR goes from 93 to 131 passing: all 13 failing MORX-3 cases and all 25 MORX-5 cases. No other results change.

A rearrangement subtable with the reverse-direction flag marks its first glyph after its last one, so it now leaves the glyphs alone. Before, it spliced with a negative range length. None of the test fonts uses one. Supporting them properly, like HarfBuzz does by reversing the buffer, would be a separate change.

Tests

• test/issues.js: four MORX-3 inputs and four MORX-5 inputs. Both tests fail on master.
• The test fonts are TestMORXThree.ttf and TestMORXFour.ttf from text-rendering-tests, added to test/data/fonttest/ under the Apache License 2.0 in test/data/fonttest/LICENSE.txt.
• npm test: 712 passing. npm run format:check passes.

This is independent of #428, #429, #430 and #432.