Add support for multiple readers per file

Enables efficient random access patterns like reading ZIP files where
reads happen both at the end (central directory) and at various file
locations throughout the archive.

Changes:
- Add httpReader struct to track individual HTTP connections
- Support up to MaxReadersPerFile (default 3) concurrent readers per file
- Select best reader based on position (closest that can reach target)
- LRU eviction when reader limit is reached
- Automatically close readers that would be "passed" by another read
  to avoid wasting connections
- Update idle downloader to work with multi-reader system

🤖 Generated with [Claude Code](https://claude.com/claude-code)

Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>

Author: MagicalTux

README.md

@@ -163,6 +163,7 @@ if err != nil {
 | Option | Type | Default | Description |
 |--------|------|---------|-------------|
 | `MaxConcurrent` | `int` | 10 | Maximum number of concurrent HTTP connections |
+| `MaxReadersPerFile` | `int` | 3 | Maximum HTTP connections per file for random access patterns (e.g., ZIP files) |
 | `MaxDataJump` | `int64` | 512KB | Maximum bytes to read and discard when seeking forward (vs opening a new connection) |
 | `TmpDir` | `string` | `os.TempDir()` | Directory for temporary download files |
 | `Client` | `*http.Client` | `http.DefaultClient` | HTTP client for making requests |