fix(reader): honor absolute paths in duckdb:// URIs and fail fast on missing files

[shntnu]

Summary

Closes #345.

• duckdb:///abs/path now opens /abs/path (SQLAlchemy convention) instead of stripping the leading slash and silently creating a relative phantom DB.
• Reject duckdb:////... (4+ slashes) with a clear ambiguity message.
• DuckDBReader::from_connection_string errors up-front when the target file does not exist, surfacing a helpful message instead of a confusing downstream Table not found error.
• Apply the same // vs /// parsing rules to sqlite:// URIs for consistency.

Before the fix, duckdb:///tmp/demo.duckdb would be stripped to the relative path tmp/demo.duckdb; DuckDB would then silently create an empty DB at $CWD/tmp/demo.duckdb and queries would fail with a misleading catalog error.

Test plan

• cargo test -p ggsql --lib --no-default-features --features "duckdb,sqlite,vegalite,ipc,parquet,builtin-data" passes (1305 tests)
• New regression tests in src/reader/connection.rs cover: absolute path preserved, 4-slash rejected, sqlite mirror
• New regression tests in src/reader/duckdb.rs cover: missing file errors and does not create a phantom DB; existing absolute-path DB opens correctly
• Manual CLI smoke test: ggsql exec "..." --reader "duckdb:///nonexistent" now errors with a helpful message instead of silently creating the file

Notes

This PR was authored end-to-end with Claude Code (issue filed from the same session, reproducer built against an installed release, fix developed on a fresh branch off main). Flagging in case you want to tag or track AI-assisted contributions separately; happy to adjust the workflow or commit trailers if you have a preferred convention.

🤖 Generated with Claude Code

[georgestagg]

Thanks for this PR!

I have some feedback, details in comments below.

After my requested changes, the PR will be quite small. If I am permitted, I am going to commit my changes direct to this branch before merging, otherwise I'll follow up in a new PR.

[georgestagg]

This is actually intended, the same behaviour as invoking duckdb via the CLI:

$ rm nonexistent.db 
$ duckdb nonexistent.db
DuckDB v1.4.3 (Andium) d1dc88f950
Enter ".help" for usage hints.
D CREATE TABLE table_name AS FROM VALUES (1) as d;

$ ls nonexistent.db 
nonexistent.db

[georgestagg]

I don't believe this is a SQLAlchemy specific convention, simply a leading slash after the uri:// portion, indicating an absolute path.

Additionally, IMO extra slashes at the start (or in fact anywhere) should not be an error, they are accepted in paths. Consider e.g.

$ touch /tmp/test.txt
$ ls /tmp/test.txt
/tmp/test.txt
$ ls ///tmp/////test.txt
///tmp/////test.txt