Add auto-language extra for code block language detection (#361)

[Darkness1521]

Closes #361

Summary

• Add auto-language extra that automatically detects the programming
  language of fenced code blocks without explicit language tags
• Uses heuristic pattern matching (no new dependencies)
• Supports 13 languages: Python, JavaScript, HTML, CSS, SQL, Bash, Java,
  Go, Rust, Ruby, PHP, JSON, YAML, C/C++

How it works

Runs before fenced-code-blocks in the processing pipeline. When a code
block has no language tag, it analyzes the content and inserts the
detected language name. The existing fenced-code-blocks and Pygments
highlighting then process it normally.

Usage

import markdown2
html = markdown2.markdown(text, extras=["fenced-code-blocks", "auto-language"])

Tests

On 24 short code snippets (the typical case for markdown code blocks):

Method	Accuracy
Our heuristic (detect_language)	24/24 = 100%
Pygments guess_lexer()	6/24 = 25%

The test suite passes with no regressions. All changes comply with the
project's contribution guidelines (PEP8, test coverage, docs updated).

Result

with auto-language:

without auto-language:

[Crozzers]

Perhaps alot of these consecutive if statements could be compressed?

Maybe something like

python = [
  [re.compile(r'^\s*(def\s+\w+\s*\(|class\s+\w+.*:)'), 5],
  [re.compile(r'^\s*(import\s+\w+|from\s+\w+\s+import\b)', re.M), 4],
  ...
]
scores['python'] = sum(score for regex, score in python.items() if regex.search(code))

[Crozzers]

This function is specific to the extra, so should probably be part of the AutoLanguage class

[Crozzers]

How many languages does guess_lexer support? If the pygments implementation is wider and more tested, perhaps offering some kind of fallback if the dependency is present?

[Crozzers]

Oh, just saw the test results you added in the PR about performance vs pygments. This is not needed then I guess