Fixes #3133.

Bug

The Markdown, reStructuredText and TiddlyWiki5 lexers all delegate the body of an embedded code block to a sub-lexer with lexer.get_tokens_unprocessed(code). That call returns indices relative to the code snippet (starting at 0), but the indices the outer lexer yields are supposed to be absolute positions within the whole input text. So every token inside a fenced/indented code block came back with a wrong offset. The MarkdownLexer even had a # FIXME: aren't the offsets wrong? comment about exactly this.

Minimal check:

from pygments.lexers.markup import MarkdownLexer
text = "intro\n```python\nx = 1\n```\n"
for index, _tok, value in MarkdownLexer().get_tokens_unprocessed(text):
    assert text[index:index+len(value)] == value  # fails on the embedded code tokens

Cause

The delegated token stream (and the do_insertions(...) output wrapping it) is never re-based by the start position of the code group, so the sub-lexer's 0-based indices leak out unchanged.

Fix

Added a small _shift_indices(tokens, offset) helper and re-based the delegated tokens by match.start(<code group>) at the four affected sites (RstLexer._handle_sourcecode, MarkdownLexer._handle_codeblock, TiddlyWiki5Lexer._handle_codeblock, TiddlyWiki5Lexer._handle_cssblock). The plain get_tokens output and the rendered HTML are byte-for-byte unchanged — only the previously incorrect positions from get_tokens_unprocessed are corrected.

Testing

Added tests/test_markup.py asserting that every token's index points at its own value in the input, for all three lexers (fails before, passes after). Full suite 5262 passed, 16 skipped; ruff check --ignore UP031 . clean.