feat(#237): blanket resource pricing via an Infracost breakdown importer

[elecnix]

Closes #237.

What

A generic path to price the static long tail without a hand-written
ResourceType handler + _METRIC_DESCRIPTORS entry per service. Infracost
already encodes extract + cost components + product filters for hundreds of
resources; this ingests its infracost breakdown --format json output and turns
each costed resource into a priced cost-model node.

This makes infra-cost-model a DAG / what-if layer on top of Infracost's pricing
library for the always-on tail — the resources otherwise forced onto hand-written
flatOverride nodes (KMS, WAF, EIP, Route53, …). Native handlers stay for the
request-path resources where the DAG derives usage from upstream flow. Per DP#9,
the import is the escape hatch, not the default: prefer a handler where one
exists.

Changes

• pricing/sources/infracost_breakdown.py — import_breakdown(json) -> {address: node}. One flatOverride node per costed resource; costComponents
  and nested subresources flatten into fixed: true metrics valued at each
  component's monthlyCost (namespaced so same-named components don't collide).
  Free resources (no priced components) skipped; provider inferred from the
  resource-type prefix.
• cli.py — import-infracost <breakdown.json> (YAML default, --json).
• tests — importer units (EC2 with an EBS subresource + ElastiCache, engine
  round-trip to the Infracost monthly total) + two CLI tests.
• README — "Blanket pricing for the long tail".

Design note

Two things block fully automatic blanket support, and this PR is honest about
both: usage derivation isn't recoverable from a resource definition (it's an
input), and AWS usagetype pricing keys aren't mechanically derivable. Infracost
has already solved both for its supported resources, so importing its breakdown
is the highest-leverage way to get breadth. What it can't do — model a resource's
usage as a function of upstream DAG flow — is exactly what the native handlers are
for, so the two compose.

Tests

pytest — full suite green (953 passed); 12 new tests.

Follow-ups (noted in the issue)

• Reconcile an imported resource that also has a native handler (prefer handler).
• Optionally wire imported nodes as DAG edges rather than always-on leaves.