Apply controllers and implement set commands during lattice expansion

[DavidSagan]

Brings lattice expansion in line with pals#237 (Controller) and pals#268 (lattice expansion), plus the set and expand_lattice sections already on pals main.

Controllers

Controllers were evaluated but never applied — control_type was read and ignored, so the lattice was untouched.

• ABSOLUTE controllers now drive the parameters their parameter pattern matches, each set to the sum over the controllers driving it, replacing the element's own value. A parameter the element does not carry is created.
• RELATIVE controllers take no part in expansion: the parameter keeps the value its element gives, and only the control expression is evaluated for the program that varies the knob later.
• Applied after branch/fork expansion and before the bookkeeper, so reference, floor and dependent parameters are computed from the driven values.
• Controllers driving another controller's variable form a hierarchy, evaluated top-down; cycles are reported. control_type defaults to ABSOLUTE and is materialized. A variable with no value is zero, and an initial value is a constant expression — neither it nor a control expression may reach outside its own controller.
• Reported: circular hierarchy, a parameter driven by both types, a parameter both controlled and expr(...)-delayed, a target matching nothing, an unknown control_type.

set and expand_lattice

Both were absent entirely. An expand_lattice node now divides the facility list and the pipeline follows the documented order:

pre-expansion sets → branch/fork expansion → ABSOLUTE controllers → bookkeeper
  → post-expansion sets → ABSOLUTE controllers → bookkeeper

• Before expand_lattice a set writes the element definitions, and only those defined earlier in the list — so every expanded copy inherits one value.
• After it, a set writes each expanded element separately, and may use computed values (SELF.s_position).
• value expressions take PARAMETER (the value being replaced) and SELF.<path> (the owning element). The compact sets: form is supported.
• An unwritten parameter reads as zero, unless a member of its linked family is written — then its value is still to be derived and reading it is an error, per the Bs1/Ks1 example.
• That same family notion lets a post-expansion set drop what it invalidated, so the second bookkeeper pass rebuilds it instead of flagging an inconsistency.

Not implemented

absolute_error/relative_error are parsed and reported rather than applied: the standard gives the error magnitude (absolute_error + relative_error * |value|) but not its distribution, and this library defers randomness everywhere else so expanded stays reproducible.

Tests

203 pass (was 174). New tests/test_sets.cpp (14 cases) and a rewritten tests/test_controllers.cpp.

🤖 Generated with Claude Code