What

Adds an optional tier field to the per-replica object in var.replicas:

variable "replicas" {
  description = "Map of NAME=> {REGION, KMS_KEY, TIER} for additional read replicas. Tier defaults to the primary instance tier when unset. Set to null to disable replica creation."
  type = map(object({
    region              = string
    encryption_key_name = optional(string)
    tier                = optional(string)
  }))
  default  = {}
  nullable = false
}

and uses it on google_sql_database_instance.replicas:

tier = coalesce(each.value.tier, var.tier)

Why

Cloud SQL allows a read replica to have a different machine tier from its primary, and this is genuinely useful:

• Cost — analytics / reporting replicas often handle a fraction of the primary's QPS and can run on a much smaller tier.
• Validation — you can move a single replica to a bigger / smaller tier first to observe behaviour before resizing the primary.
• Regional sizing — a replica in a cheaper or more constrained region may need a different tier than the primary.

Today the module hardcodes every replica to var.tier, so the only workaround is to manage replicas outside the module.

Compatibility

• The new field is optional(string), so existing callers that pass replicas = { ... } without a tier key are unaffected: coalesce(null, var.tier) returns var.tier, which is exactly today's behaviour.
• The field lives on the per-replica object, so different replicas can have different tiers (or inherit).
• Does not touch the primary's tier handling.

Does not conflict with the sibling retain-backups change (#27) — that one touches var.tier-adjacent variables.tf only to add a separate variable and edits the settings {} block to wire it in; this PR's only main.tf change is the tier = coalesce(...) line, and its only variables.tf change is the replicas type/description.