string_view: fix HasWord false negative after a non-boundary partial match

[EylonKrause]

Description

CpuFeatures_StringView_HasWord detects a separator-delimited word in a line (used to find CPU features in /proc/cpuinfo flag lists). It searches for the word in remainder — which advances every loop iteration — but runs the separator-boundary checks by slicing before/after from the original line using that remainder-relative index:

const int index_of_word = CpuFeatures_StringView_IndexOf(remainder, word);
...
const StringView before = CpuFeatures_StringView_KeepFront(line, index_of_word);
const StringView after  = CpuFeatures_StringView_PopFront(line, index_of_word + word.size);
...
remainder = CpuFeatures_StringView_PopFront(remainder, index_of_word + word.size);

On the first iteration remainder == line, so the index is correct. But once an earlier, non-boundary partial occurrence advances remainder, index_of_word is relative to the shorter remainder while still being applied to the full line — so the boundary checks examine the wrong characters and the function returns a false negative for a word that really is present as a token.

Example: HasWord("foox foo", "foo", ' ') returns false even though foo is the second token. This affects any feature whose name first appears as a non-boundary substring of an earlier token and then again as its own token (e.g. MIPS/LoongArch flag lists such as ... mips32r2 mips32 ... searching for mips32).

Fix

Convert the remainder-relative index to an absolute index into line before slicing (remainder is always a suffix of line, so its offset is line.size - remainder.size). This keeps the boundary checks anchored to the full line.

Note: naively switching the slices to remainder instead would introduce a false positive at remainder-start boundaries — e.g. HasWord("foofoo bar", "foo", ' ') would then wrongly match. The absolute-index form avoids that.

Testing

Extended the existing StringViewTest.CpuFeatures_StringView_HasWord test:

• Before the fix: HasWord("foox foo","foo",' ') and HasWord("xvme vme","vme",' ') return false → the test FAILS.
• After the fix: both return true, the no-false-positive guard (HasWord("foofoo bar","foo",' ') → false) holds, and the full suite is green under ASan/UBSan (ctest: 4/4; string_view_test: 17/17).