Off-by-one in SetPosition/GetTextSize truncates block selections by one character

[Egbert-Azure]

UpperCaseBlock, LowerCaseBlock and SentenceCaseBlock fail to convert the final character of the block. hello world becomes HELLO WORLd; σιγμα becomes σιγμΑ.

Pre-existing, not a regression. git blame shows the only recent change in these functions is the one-line end+1 → end fix (f3cce7e, fab4b20, e8b2456, 2026-09-06); everything else dates to d47b062 and 42dd4cb. The bug was previously masked: with +1, GetBlockText always passed a trailing \r to the converter, and case conversion is a no-op on \r, so the sacrificial last character hid the defect. Removing +1 passes real block text and exposes it.

The defect appears to be in the conversion path — cDocument::UpperCase()/LowerCase()/SentenceCase() (document.cpp:4197-4224) wrap NormalizeToUTF32() → unicode::to_uppercase() → to_nfc() → utf8::encode(). Not yet pinned to a specific line; likely inside the third-party unicode library rather than src/.

Note that TEST_CASE("Lowercase") at test-document.cpp:2663 cannot detect this — its fixture already ends in a lowercase letter.

Failing tests (left red deliberately, in test-guieditor.cpp): UpperCaseBlock, LowerCaseBlock, SentenceCaseBlock, and the case-conversion integration test.

[Egbert-Azure]

Diagnosis revised. This is not a defect in the case-conversion path — the conversion functions are correct.

The truncation comes from a pre-existing off-by-one in cDocument::SetPosition()'s bounds clamp (or in GetTextSize()'s count), which shortens block selections by one character. Case conversion only exposes it: with the old end+1, the sacrificial trailing character absorbed the error, so it was invisible.

Fix site is SetPosition()/GetTextSize(), not document.cpp:4197-4224. Retitled and re-scoped accordingly.

Same family as #15 and the ^KZ boundary defects — capture a position, mutate or clamp, reuse without recomputation.