Add wide table CSV data load

[hbarthels]

What

Adds a table form to csv_data for loading tabular CSV data into a single output relation with schema key=[T1, T2, ..., TN], value=[] — no synthesized row ID in the key.

The new S-expression syntax:

; per-column mode (unchanged)
(columns
  (column "src" :col_src [UINT128 INT])
  (column "dst" :col_dst [UINT128 INT]))

; new table mode — no row ID in output
(table :edges ["src" "dst"] [INT INT])

Changes

Proto (logic.proto)

• New CSVTarget message: target_id (RelationId), column_names (repeated string), types (repeated Type)
• optional CSVTarget target = 5 on CSVData; mutually exclusive with the existing columns field

Grammar (grammar.y)

• New csv_table rule: "(" "table" relation_id "[" STRING* "]" "[" type* "]" ")"
• csv_data rewritten to a single alternative with gnf_columns? csv_table? — both optional and distinguishable by LL(2) lookahead ("columns" vs "table"), avoiding the LL(3) conflict that would arise from two alternatives sharing the long csvlocator csv_config prefix
• Construct/deconstruct helper functions for the two modes

Generated files

• Parsers, pretty-printers, and proto stubs regenerated for Python, Julia, and Go

Tests

• tests/lqp/csv_table.lqp — four table-mode variants (int/int edges, string/float scores, single-column, custom delimiter) with binary and snapshot artifacts

Reviewer notes

• The __slots__ changes in transactions_pb2.pyi and other .pyi stubs are a pre-existing version drift between the committed stubs and the current buf pyi plugin (1.56.0); they are not caused by the proto changes here.