feat(sources): Source bundles for debug files

[jan-auer]

This PR contains two major additions to the debuginfo crate:

• DebugSession::source_by_path now exposes source code contents that are stored within object files. However, this is not implemented for any of the existing object file formats yet. The information whether an object file might even contain sources is exposed via ObjectLike::has_source.

• A new object file format, SourceBundle, allows for platform-independent bundling of source context. It is a ZIP archive containing all source files referenced by an object file. Currently, this is derived by walking the debug information in the source file.

SourceBundle

Source bundles are a ZIP zip files with a well-defined internal structure. First it contains a binary header, followed by a ZIP stream:

static BUNDLE_MAGIC: [u8; 4] = *b"SYSB";

#[repr(C, packed)]
struct SourceBundleHeader {
    magic: [u8; 4], // Magic bytes header.
    version: u32,   // Version of the bundle.
}

The internal structure of the ZIP file is as follows:

manifest.json
files/
  index.js
  index.js.map

The manifest describes the entire contents of the bundle. Files that are not listed in the manifest are skipped. The manifest for the above bundle would look like this:

{
    "files": {
        "files/C/Users/test/myapp/main.cpp": {
            "path": "C:\\Users\\test\\myapp\\main.cpp",
            "type": "source"
        },
        "files/usr/local/include/somelib/somelib.h": {
            "path": "/usr/local/include/somelib/somelib.h",
            "type": "source",
        }
    },
    // ... arbitrary other attributes
}

Path Canonicalization

Paths in the source bundle are stored in a canonical way. This makes it easier to match them from different platforms. This functionality is implemented in symbolic_common::path::clean_path. Roughly, it performs the following actions:

• Replace backslashes with forward slashes
• Remove . path components
• Pop the parent directory for .. path components.
• Align to the root directory (e.g. drive letter, leading slash)
• Remove double slashes

For instance, the path C:\Users\..\..\Program Files\\Hello becomes C:\Program Files\Hello.

TODO

• Speed up file extraction. At the moment, all line records need to be walked. Instead, there could be something like DebugSession::file_names() -> Iterator<&str> that just emits all source file names.
• Read source contents from the llvm DWARF extension?
• Read source contents from PDBs?