Investigate prebuild create vs claim build times in dogfood

## Description

In our dogfood environment, the “Write Coder on Coder” template has prebuilds enabled for certain presets. We recently introduced new metrics for prebuild creation time (building into the pool) and prebuild claim time https://github.com/coder/coder/pull/19503, and added them to our observability stack.

These metrics show that claim durations sometimes exceed creation by about 1–2 seconds. Because prebuilds maintain a ready-to-use pool to reduce workspace start time, claims should be consistently faster than creations.

We already use a Terraform `lifecycle` block with `ignore_changes` as recommended in the [docs](https://coder.com/docs/admin/templates/extending-templates/prebuilt-workspaces#preventing-resource-replacement), yet this pattern suggests Terraform may still be replacing or re-provisioning resources during claim, or the claim path is performing extra work.

For reference, see the attached screenshot comparing average build times for a prebuild creation versus a claim on the Falkenstein preset in our dogfood instance.

## Investigation scope

Investigate the source of the additional claim-path latency relative to the create path.

Commencez par la création des prebuilds et les métriques de claim dans la stack d’observabilité, en comparant les données de dogfood de Falkenstein et le bloc lifecycle de Terraform avec ignore_changes décrit dans la documentation liée. Suivez les chemins de create et de claim afin d’identifier l’origine des 1–2 secondes supplémentaires ; le travail est terminé lorsque la latence supplémentaire de claim est expliquée et que tout remplacement ou reprovisionnement Terraform est confirmé ou écarté.

Rédigé par le modèle d'indexation à partir du texte de l'issue.