fix(ci): BDD In-Network ENOSPC during tox uv sync after #1613/#1634

[mkostromin-sigma]

Summary

main CI is red after #1613 / #1634. Quality Gate itself is green on the tip; BDD In-Network (e2e_rest) fails twice in a row while uv sync --group dev installs into the tox_data volume:

error: Failed to install: ... No space left on device (os error 28)

Seen on:

• https://github.com/prebid/salesagent/actions/runs/29512105304 (merge of ci: ratchet mypy --check-untyped-defs error count (#1611) #1634 + failed rerun)

#1534's main CI was green; the ratchet merges did not change compose/tox deps, but the in-network job now OOMs the Docker disk consistently on ubuntu-latest (image build + full /opt/venv + second full tox env under /app/.tox).

Fix

Harden bdd-in-network in .github/workflows/ci.yml:

1. Free runner disk before compose build (remove unused preinstalled toolchains + docker builder prune)
2. Cap PGDATA_TMPFS_SIZE for this serial job (default 10g is for heavy xdist; e2e_rest here is serial)

Relationships

• Parent: test(bdd): retire the e2e_rest xfail ledger via transport-aware harness setup #1418 (in-network e2e_rest CI gate)
• Related: ci: ratchet ruff C901/PLR0912/PLR0915 via .ruff-complexity-baseline (#1610) #1613, ci: ratchet mypy --check-untyped-defs error count (#1611) #1634 (triggered the red-main window; Quality Gate path already corrected in ci: ratchet mypy --check-untyped-defs error count (#1611) #1634)
• Blocks: anything waiting on green main

[mkostromin-sigma]

Fix up in a PR (disk reclaim + PGDATA_TMPFS_SIZE=2g + arch guard). Parent: #1418.

Root cause on the red main tip after #1634: ENOSPC during tox uv sync into tox_data, not the ratchet count logic (Quality Gate already green).

[mkostromin-sigma]

PR: #1663