Do not work with terraform-aws-modules/rds-aurora module

[igorozerskii-pa]

Changing cluster_instance_class from db.r7g.large to db.r7g.2xlarge via terraform-aws-modules/rds-aurora/aws produced no change in the reported cost.
The catalogue entry (resources/aws/aws_rds_cluster_instance.toml) refers to instance_class, but the rds-aurora module assigns it using try(coalesce(each.value.instance_class, var.cluster_instance_class), null).
c3x cannot follow that level of indirection, so it falls back to the catalogue default of db.r6g.large. When priced directly, it is correct ($1,227.13 vs $288.35 per month).

Practical effect: the delta column will miss instance class changes made through registry module variables. The report footer states that totals are a lower bound, and the --show-skipped warnings appear in the collapsed per-stack job log groups.

[andreacappelletti97]

Root cause found: c3x evaluated each variable's value but ignored its declared type constraint, so optional() object attributes were never materialised. The instances input is typed map(object({ instance_class = optional(string), ... })), and Terraform converts { one = {} } into { one = { instance_class = null } }. c3x left it an empty object, so each.value.instance_class errored on the missing attribute, try(coalesce(...), null) swallowed the whole expression to null, and the resource dropped back to the catalogue default class.

Fixed in #54 by applying the declared type constraint (the way Terraform decodes it) for both root variables and child-module inputs. Changing cluster_instance_class now flows through the coalesce correctly, both directly and via the module ($246.74 to $1,049.01 in my repro). Regression tests added; full suite green.

One boundary worth noting: this covers optional()-typed inputs. An input typed any would still not expose the attribute, but that fails in real Terraform too, so matching Terraform there is intentional.