closes #118

summary

this fixes the join behavior when a requested join key does not exist. instead of silently hashing an empty key set and producing cartesian style output, joins now fail fast with the existing columnnotfoundexception.

what changed

• added a shared validation helper in join.hs that checks requested join keys against dataframe column indices before hashes are computed
• routed innerjoin, leftjoin, rightjoin, fullouterjoin, and buildhashcolumn through that validation path
• kept the public join signatures unchanged
• preserved the correct rightjoin callpoint by routing it through an internal leftjoin helper with an explicit callpoint
• added regression tests for missing join keys across inner, left, right, and full outer joins
• updated subset.hs to use randomgen and split so the test suite builds under the current nix toolchain

behavior

before this change, a missing join column could fall through to hashing zero key columns, which made every row hash the same and could expand into cartesian style output. after this change, the join fails immediately with the missing column name and the existing available column context.

verification

• source /nix/var/nix/profiles/default/etc/profile.d/nix-daemon.sh
• nix --extra-experimental-features 'nix-command flakes' develop --no-write-lock-file -c cabal test test:tests --test-show-details=direct --test-options='-p Operations.Join'\n- nix --extra-experimental-features 'nix-command flakes' develop --no-write-lock-file -c cabal test test:tests --test-show-details=direct\n\n## results\n- full tests suite passed\n- 445 hunit cases passed\n- 0 failures\n- 0 errors\n- all quickcheck properties passed