Add powermglot: Power Query M parser + transpiler; integrate into pbi2dbr

[Copilot]

Implements a proper Power Query M parser (powermglot) as a new workspace library and replaces pbi2dbr's regex-based M expression handling with it.

powermglot — new workspace library

A recursive-descent M parser and sqlglot-backed SQL transpiler for let…in chain expressions.

Architecture:

• lexer.py — full M tokeniser: keywords, #"quoted identifiers", "" escaped strings, all operators
• ast_nodes.py — dataclass AST (LetExpr, CallExpr, NavExpr, EachExpr, BinaryOpExpr, …)
• parser.py — recursive-descent parser: nested let, each, if/then/else, dotted function names, {[Name=…]}[Data] navigation chains
• transpiler.py — walks the let binding graph (following variable references) and emits a sqlglot SELECT

Supported M → SQL mappings:

M pattern	SQL
Connector + {[Name=…]}[Data] chain (2–3 levels)	FROM catalog.schema.table
Table.SelectRows(t, each predicate)	WHERE (and/or/not)
Table.SelectColumns / RenameColumns / AddColumn / RemoveColumns	Column projection + aliases
Table.Group	GROUP BY + aggregations
Table.NestedJoin / Table.Join	JOIN
Value.NativeQuery(src, "sql")	Subquery passthrough

from powermglot import m_to_sql, parse_m_source

sql = m_to_sql("""
    let
        Source = Databricks.Catalogs("host", "443", [Catalog="prod"]),
        db = Source{[Name="pbi"]}[Data],
        orders = db{[Name="orders"]}[Data],
        filtered = Table.SelectRows(orders, each [status] = "Active")
    in filtered
""", dialect="spark")
# SELECT * FROM prod.pbi.orders WHERE status = 'Active'

parse_m_source — extraction API for embedders

A safe, never-raising entry point that returns a MSourceInfo(source_ref, native_sql, filter_sql) without generating SQL — designed for use by data-source extractors.

from powermglot import parse_m_source

info = parse_m_source(m_expr)
# info.source_ref  → "prod.pbi.orders"
# info.filter_sql  → "status = 'Active'"
# info.native_sql  → None

pbi2dbr integration

pbi2dbr now depends on powermglot (workspace dependency). The three M-resolution functions in extractor.py use parse_m_source as primary with the existing regex logic as fallback for patterns powermglot cannot yet parse (e.g. {[Item="x"][Kind="Table"]} navigation style):

• PbixExtractor._resolve_uc_ref() — powermglot resolves the table ref; applies default catalog/schema to fill in missing parts
• _extract_filter_expr() — powermglot extracts Table.SelectRows predicates via the AST; handles nested let chains and variable references the regex cannot
• _extract_native_query_sql() — powermglot extracts Value.NativeQuery SQL directly from the parsed AST

---

⚡ Quickly spin up Copilot coding agent tasks from anywhere on your macOS or Windows machine with Raycast.