agent: Improve Gemini support in the edit_file tool (#31116)

This change improves `eval_extract_handle_command_output` results for
all models:

Model                       | Pass rate before | Pass rate after
----------------------------|------------------|----------------
claude-3.7-sonnet           |  0.96            | 0.98
gemini-2.5-pro              |  0.35            | 0.86
gpt-4.1                     |  0.81            | 1.00

Part of this improvement comes from more robust evaluation, which now
accepts multiple possible outcomes. Another part is from the prompt
adaptation: addressing common Gemini failure modes, adding a few-shot
example, and, in the final commit, auto-rewriting instructions for
clarity and conciseness.

This change still needs validation from larger end-to-end evals.


Release Notes:

- N/A

crates/language/Cargo.toml

@@ -66,6 +66,7 @@ tree-sitter.workspace = true
 unicase = "2.6"
 util.workspace = true
 workspace-hack.workspace = true
+diffy = "0.4.2"
 
 [dev-dependencies]
 collections = { workspace = true, features = ["test-support"] }

crates/language/src/language.rs

@@ -65,7 +65,9 @@ use std::{num::NonZeroU32, sync::OnceLock};
 use syntax_map::{QueryCursorHandle, SyntaxSnapshot};
 use task::RunnableTag;
 pub use task_context::{ContextProvider, RunnableRange};
-pub use text_diff::{DiffOptions, line_diff, text_diff, text_diff_with_options, unified_diff};
+pub use text_diff::{
+    DiffOptions, apply_diff_patch, line_diff, text_diff, text_diff_with_options, unified_diff,
+};
 use theme::SyntaxTheme;
 pub use toolchain::{LanguageToolchainStore, Toolchain, ToolchainList, ToolchainLister};
 use tree_sitter::{self, Query, QueryCursor, WasmStore, wasmtime};

crates/language/src/text_diff.rs

@@ -1,4 +1,5 @@
 use crate::{CharClassifier, CharKind, LanguageScope};
+use anyhow::{Context, anyhow};
 use imara_diff::{
     Algorithm, UnifiedDiffBuilder, diff,
     intern::{InternedInput, Token},
@@ -119,6 +120,12 @@ pub fn text_diff_with_options(
     edits
 }
 
+pub fn apply_diff_patch(base_text: &str, patch: &str) -> Result<String, anyhow::Error> {
+    let patch = diffy::Patch::from_str(patch).context("Failed to parse patch")?;
+    let result = diffy::apply(base_text, &patch);
+    result.map_err(|err| anyhow!(err))
+}
+
 fn should_perform_word_diff_within_hunk(
     old_row_range: &Range<u32>,
     old_byte_range: &Range<usize>,
@@ -270,4 +277,12 @@ mod tests {
             ]
         );
     }
+
+    #[test]
+    fn test_apply_diff_patch() {
+        let old_text = "one two\nthree four five\nsix seven eight nine\nten\n";
+        let new_text = "one two\nthree FOUR five\nsix SEVEN eight nine\nten\nELEVEN\n";
+        let patch = unified_diff(old_text, new_text);
+        assert_eq!(apply_diff_patch(old_text, &patch).unwrap(), new_text);
+    }
 }