Fix MapAt for parts that are lists, other heads and several positions

[tombouctou]

MapAt gave wrong results whenever the part at the position was itself a list, which is common with matrices:

MapAt[f, {{1, 0}, {0, 1}}, {2}]            (* gave {0, f[1]};  WMA: {{1, 0}, f[{0, 1}]} *)
MapAt[0 # &, {{1, 0}, {0, 1}}, {{2}}]      (* gave {0, 0};     WMA: {{1, 0}, {0, 0}} *)
MapAt[f, {{1, 0}, {0, 1}}, {{1}, {2}}]     (* gave {f[1], f[0]}; WMA: {f[{1, 0}], f[{0, 1}]} *)

With a one-index position, the code applied f inside the part at that position and returned that part's elements as the whole result. Also:

MapAt[g, h[a, b], 2]                       (* gave {a, g[b]};  WMA: h[a, g[b]] *)
MapAt[g, {a, k[b, c]}, {2, 1}]             (* gave MapAt::partw; WMA: {a, k[g[b], c]} *)
MapAt[g, {{a, b}, {c, d}}, {{1, 2}, {2}}]  (* gave {c, g[d]};  WMA: {{a, g[b]}, g[{c, d}]} *)

This replaces the position handling in eval_MapAt with one recursive function that walks the position and rebuilds each level with its own head. Error messages (MapAt::partw, MapAt::psl), negative positions, associations, and the existing handling of rules are unchanged. The doc examples still pass.

Tests: new cases in test/builtin/functional/test_apply_fns_to_lists.py. pytest and the doctests pass.

Found while running Strickland's LambdaAlgebra.m, which builds a projection matrix with MapAt[0 # &, IdentityMatrix[n], {{i}, ...}].

🤖 Generated with Claude Code