diff kitten: Speedup patch parsing by working with bytes rather than unicode

Also change the line split algorithm to only split on \n, \r and \r\n.
This is hopefully closer to what git/diff generate in their patch files.
I cant find any documentation specifying this however.

Fixes #6052
Fixes #6092

kittens/diff/patch.py

@@ -8,6 +8,8 @@ import shutil
 import subprocess
 from typing import Dict, Iterator, List, Optional, Sequence, Tuple, Union
 
+from kitty.fast_data_types import splitlines_like_git
+
 from . import global_data
 from .collect import lines_for_path
 from .diff_speedup import changed_center
@@ -37,7 +39,7 @@ def set_diff_command(opt: str) -> None:
     global_data.cmd = cmd
 
 
-def run_diff(file1: str, file2: str, context: int = 3) -> Tuple[bool, Union[int, bool], str]:
+def run_diff(file1: str, file2: str, context: int = 3) -> Tuple[bool, Union[int, bool], bytes]:
     # returns: ok, is_different, patch
     cmd = shlex.split(global_data.cmd.replace('_CONTEXT_', str(context)))
     # we resolve symlinks because git diff does not follow symlinks, while diff
@@ -53,8 +55,8 @@ def run_diff(file1: str, file2: str, context: int = 3) -> Tuple[bool, Union[int,
     returncode = p.wait()
     worker_processes.remove(p.pid)
     if returncode in (0, 1):
-        return True, returncode == 1, stdout.decode('utf-8')
-    return False, returncode, stderr.decode('utf-8')
+        return True, returncode == 1, stdout
+    return False, returncode, stderr
 
 
 class Chunk:
@@ -158,19 +160,19 @@ class Hunk:
             c.finalize()
 
 
-def parse_range(x: str) -> Tuple[int, int]:
-    parts = x[1:].split(',', 1)
+def parse_range(x: bytes) -> Tuple[int, int]:
+    parts = x[1:].split(b',', 1)
     start = abs(int(parts[0]))
     count = 1 if len(parts) < 2 else int(parts[1])
     return start, count
 
 
-def parse_hunk_header(line: str) -> Hunk:
-    parts: Tuple[str, ...] = tuple(filter(None, line.split('@@', 2)))
+def parse_hunk_header(line: bytes) -> Hunk:
+    parts: Tuple[bytes, ...] = tuple(filter(None, line.split(b'@@', 2)))
     linespec = parts[0].strip()
     title = ''
     if len(parts) == 2:
-        title = parts[1].strip()
+        title = parts[1].strip().decode('utf-8', 'replace')
     left, right = map(parse_range, linespec.split())
     return Hunk(title, left, right)
 
@@ -190,25 +192,30 @@ class Patch:
         return len(self.all_hunks)
 
 
-def parse_patch(raw: str) -> Patch:
+def parse_patch(raw: bytes) -> Patch:
     all_hunks = []
     current_hunk = None
-    for line in raw.splitlines():
-        if line.startswith('@@ '):
-            current_hunk = parse_hunk_header(line)
+    plus, minus, backslash = map(ord, '+-\\')
+
+    def parse_line(line: memoryview) -> None:
+        nonlocal current_hunk
+        if line[:3] == b'@@ ':
+            current_hunk = parse_hunk_header(bytes(line))
             all_hunks.append(current_hunk)
         else:
             if current_hunk is None:
-                continue
-            q = line[0] if line else ''
-            if q == '+':
+                return
+            q:int = line[0] if len(line) > 0 else 0
+            if q == plus:
                 current_hunk.add_line()
-            elif q == '-':
+            elif q == minus:
                 current_hunk.remove_line()
-            elif q == '\\':
-                continue
+            elif q == backslash:
+                return
             else:
                 current_hunk.context_line()
+
+    splitlines_like_git(raw, parse_line)
     for h in all_hunks:
         h.finalize()
     return Patch(all_hunks)
@@ -244,7 +251,7 @@ class Differ:
             except Exception as e:
                 return f'Running git diff for {left_path} vs. {right_path} generated an exception: {e}'
             if not ok:
-                return f'{output}\nRunning git diff for {left_path} vs. {right_path} failed'
+                return f'{output.decode("utf-8", "replace")}\nRunning git diff for {left_path} vs. {right_path} failed'
             left_lines = lines_for_path(left_path)
             right_lines = lines_for_path(right_path)
             try: