0.3.0

search and read_file share 1-based line numbers

• Breaking: read_file(path, offset=0, limit) is now read_file(path, start_line=1, limit), and start_line is 1-based. offset was 0-based while search reports 1-based
  lines, so passing a hit's line as offset started the read one line after the match.
• search takes context (0–10). Each hit then also carries before and after, the
  bounded text of up to that many neighbouring lines, so a match can be seen in one call.
  Context is read the same way for ripgrep and the Python fallback, so both give identical
  results; a neighbour that is itself a match still appears as context.

edit_file, glob filtering, search output modes, numbered reads

• New edit_file(path, old, new, replace_all=False) tool in the files bundle: replaces exact
  text, so changing part of a file no longer means resending all of it through write_file.
  old must be unique unless replace_all is set; a miss or an ambiguous match is an error
  that names the lines involved (or a near match differing only in case or whitespace) and
  leaves the file untouched. The write is atomic and keeps the file's line endings and
  permissions. It refuses the directories write_file refuses, binary and non-UTF-8 files,
  and files over 10 MiB; its reply shows the edited lines, numbered and bounded.
• list_files takes a glob pattern matched against paths relative to path (* within one
  folder, ** across folders, e.g. **/*.csv).
• search takes output: "hits" (default, unchanged), "files" (each matching file once)
  or "count" (matching lines per file). For the file outputs max_matches caps files, not
  lines. Both backends give the same results.
• read_file takes line_numbers, prefixing each line with its number and a tab.