[caching] save/load source hierarchy and extracted log statements

[tstack]

Scanning the source hierarchy and extracting log statements is pretty costly, so we need to cache what we found between runs. This change writes a separate file for every source root. The file contains a header that is a JSON-line with interesting metadata that a user might find helpful. The body of the file is a bincode serialization of the SourceTree object. The CLI will try to load from the cache on startup, check for any changes in the source tree, and then update the cache, if necessary.

I was originally looking at using SQLite for the cache, but it's hard to write blobs of unknown size to a table. This approach seems to work since the header can easily be read separately from the main content to determine if the file is valid or not. I also tried using the postcard
serializer, but had trouble getting it to work. The bincode one seems to just work.

Files:

• Cargo.toml: Add some more dependencies:
  • bincode for serialization
  • directories for finding the user's platform-specific cache directory
  • serde_regex for serializing/deserializing Regex
  • sha2 for producing a hash used for the cache file names
  • tempfile for creating the cache file
  • walkdir for walking the cache dir in the tests
• cache-header-v1.json: The schema for the cache entry header
• lib.rs: Add cache stuff
• main.rs: Add a footer to the help to mention where the cache is. Load the statements from the cache and save for future use.
• source_hier.rs: Derive Deserialize on various structs.
• common_settings/mod.rs: Move helper module to its own directory.
• source_ref.rs: Serialize/deserialize the Regex directly so the deserialize can fail if the regex string is invalid. Add separate pattern_str to cache the string version of the pattern.

[ttiimm]

Does ProjectDirs need TEMP or TMP set to work on windows? That was one recommendation by the coding assistant on how to fix this...