Name-plus-module matching of a dump's instance tree picks same-module siblings

Suppose you map a simulator's elaborated hierarchy (for example a VCS KDB) onto another tool's elaboration (for example slang) by walking the instance chain top-down and accepting the first level where some candidate subtree has the same relative names and the same modules. When the two elaborations differ, that rule silently maps an instance to a sibling of the same module: u_fifo[0] becomes u_fifo[1], g[3] becomes g[0] after a parameter override, u_dut becomes a spare instance. The true counterpart failed the module check deeper down, so the search moved on to another candidate. Two fixes: once a candidate matches modules down to some depth and then fails, report 'other module' instead of searching further; and below a matched counterpart, only accept the same relative name. A perturbation sweep catches this where unperturbed sweeps never do (0 wrong unperturbed, 47-541 wrong rows per mode perturbed). Perturb the reference: remove a subtree, change one instance's module, rename one instance.