fix: replace unsafe byte indexing in parse_value with .get() + error return

Direct byte indexing (source[span.0..span.1] and source[..span.0]) panics
if yamlpath returns a span misaligned to UTF-8 boundaries, crashing the
Python process. Replace with .get() to return a recoverable PyValueError,
matching the existing safe pattern in extract().

Closes #44

Author: nathanjmcdougall

src/document.rs

@@ -94,11 +94,20 @@ impl PyDocument {
             match self.inner.query_exact(&r) {
                 Ok(Some(feature)) => {
                     let span = feature.location.byte_span;
-                    let raw = &source[span.0..span.1];
+                    let raw = source.get(span.0..span.1).ok_or_else(|| {
+                        PyErr::new::<pyo3::exceptions::PyValueError, _>(
+                            "Feature span is not valid in source",
+                        )
+                    })?;
                     // Calculate the column offset (in bytes) of the value
                     // start relative to the beginning of its line, so we can
                     // dedent continuation lines.
-                    let line_start = source[..span.0].rfind('\n').map(|nl| nl + 1).unwrap_or(0);
+                    let prefix = source.get(..span.0).ok_or_else(|| {
+                        PyErr::new::<pyo3::exceptions::PyValueError, _>(
+                            "Feature span start is not valid in source",
+                        )
+                    })?;
+                    let line_start = prefix.rfind('\n').map(|nl| nl + 1).unwrap_or(0);
                     let col = span.0 - line_start;
                     if col == 0 {
                         raw.to_string()