Add support for custom dictionaries

[Zenous07]

What this does

Adds support for registering custom dictionaries in lithiumpatch, on top of the existing built-in ones (edgedict, webster1913). Users can now convert any dictionary they own into a simple JSON format and plug it into the build — no need to touch the core patching logic.

This is implemented as a new dict/custom package that:

• Embeds one or more JSON dictionary files using //go:embed.
• Registers each one with the existing dict.Register(name, priority, parseFunc) API — the same mechanism the built-in dictionaries already use.
• Requires only a single blank import in main.go (_ "github.com/pgaskin/lithiumpatch/dict/custom") to activate.

How to use it

1. Convert your source dictionary (StarDict, CSV, or anything else) into a JSON array matching the existing dict.Entry struct — each entry has Terms, Name, MeaningGroups (with Meanings[].Text etc.), and Source. No changes were made to dict.Entry itself.
2. Place the resulting .json file in dict/custom/.
3. Add a //go:embed + dict.Register(...) block for it in dict/custom/custom.go — a couple of lines per dictionary.
4. Build as usual: go run . app/Lithium_0.24.5.apk. The dictionary shows up in the reader's lookup feature alongside the built-in ones.

Full details, including a worked StarDict conversion example and a CSV conversion example, are documented in the new "Custom Dictionaries" section added to the README.

Backward compatibility

This change is fully additive and backward-compatible:

• No existing dictionary code, structs, or registration logic was modified.
• Built-in dictionaries (edgedict, webster1913) continue to work exactly as before if the user doesn't touch dict/custom.
• If a user doesn't add any custom dictionaries or doesn't import the custom package, behavior is identical to the current master branch.

Testing

Built and tested locally with go run . app/Lithium_0.24.5.apk. Confirmed the custom dictionary registered via dict/custom appears correctly in the app's word-lookup UI alongside the built-in dictionaries, with no regressions to existing dictionary behavior.