Remove fisher(), which never returned a p-value

[zhiaos]

Addresses #30. Not closing it yet — leaving that to @iblacksand, who confirmed the method was abandoned in favour of stouffer.

fisher() returned a chi-square density on 2^(k-1) degrees of freedom rather than an upper-tail probability on 2k. The two defects compound into an inversion: three p-values of 0.999 combined to 0.0015, five to 7.7e-21. Details and the measured table are in #30.

Removes the function, the MetaAnalysisMethod::Fisher variant and its two match arms. Removal rather than repair follows @iblacksand's note on the issue: nothing calls it — webgestaltpy uses stouffer, and WebGestaltR asked Rust for "fisher" but overwrote the result with poolr::stouffer — so a corrected implementation would have no users, while a wrong one is a trap for anyone who later removes the "redundant" R computation.

The Continuous import went with it; it was only needed for the .pdf() call.

Breaking for anyone matching on MetaAnalysisMethod, which now has a single variant. Worth a 0.4.0 rather than 0.3.3 whenever this is published — the seed field added in #31 is breaking for struct-literal construction of GSEAConfig for the same reason.

WebGestaltR is updated in a companion PR so its bridge no longer names Fisher; that lands first, so nothing breaks on the next re-vendor.

cargo test --all passes (5 tests), and cargo vendor still resolves cleanly — the check #34 exists to remind us about.