fix(codegen): scope-unique block var members + collection lvalue aliasing

[luisleo526]

Two independent Pine reference/scope correctness bugs, each proven by hand-edit and validated by the full gate + an 87-strategy behavioral sweep (0 regressions).

BUG 1 — block-scoped var name collision

Block-scoped var/varip declarations were keyed by raw name and deduped by name, so two same-named vars in sibling non-global scopes shared one C++ member.

analyzer/base.py::_visit_VarDecl now assigns a scope-unique member {name}__blk{seq} when a block-scoped var's raw name already belongs to a different enclosing block, records the rename in AnalyzerContext.block_var_renames (keyed by block node via a new _block_node_stack), and codegen activates it through the existing _active_var_remap plumbing.

→ egoigor1976-1-trendline (two var bool valid, upper vs lower trendline): 44.2% → 100.0% (price-exact).

BUG 2 — collection lvalue aliasing value-copied

Pine arrays/maps/matrices are reference types, but a local bound to an existing collection lvalue (incl. a ternary/switch of collection lvalues) was value-copied, so mutations never reached the member. Extended the existing UDT-lvalue-alias mechanism to collections (types.py helpers + visit_stmt.py emits T& x = rhs; for a qualifying mutated local).

→ jevondijefferson-big-breakout (orderBlocks aliased to internal/swing member via ternary then unshift/remove): 36.1% → 76.6% (price-exact 99.4%).

Validation

	result
full gate PINEFORGE_ENGINE_INCLUDE=... pytest	1390 passed, 1 skipped, 0 failed (1386 + 4 teeth tests; 254-corpus compile green)
broad 87-strategy sweep	exactly 2 movers (the targets), 0 regressions; 14 unrelated multi-helper strategies byte-identical
teeth	both regression tests fail on the unpatched tree; 2 no-op guards pin "no-op unless triggered"

Both fixes are no-ops unless their construct is present → corpus regeneration stays safe.

🤖 Generated with Claude Code