feat(check-endpoint.py): add support for body capture (AH-202608128171)

[bytebeast]

Refs: AH-202608128171

What does this PR do?

Why

Type of change

• fix - bug fix (patch release)
• feat - new flag/column/output mode (minor release)
• perf - performance improvement (patch release)
• refactor / docs / chore / ci - no user-facing behavior change (no
  release)
• Breaking change (! after the type, or BREAKING CHANGE: in the commit
  body - triggers a major release)

Commit message format

This repo computes version bumps and the CHANGELOG entry from your commit
messages (.github/scripts/release.py), not from this PR description. Please
use conventional commits: type(scope): description, e.g.
fix(timeout): show <TO> marker when BODY_DL fails. Only feat/fix/perf
(or a breaking change) trigger a release; docs/chore/refactor/ci don't.
Ticket refs go in one of the three recognized spots (legacy
PROJ-123 fix(...): ... prefix, trailing (PROJ-123), or a Refs:/Closes:
footer) - see the module docstring in .github/scripts/release.py for the full
list.

Do NOT hand-edit for this PR

• APP_VERSION in check-endpoint.py
• version in pyproject.toml
• APP_VERSION in contrib/check-endpoint-exporter/check-endpoint.py
• CHANGELOG.md

These are written automatically by the chore(release): PR the release workflow
opens after your PR merges, based on the conventional-commit types above. A
manual bump here will conflict with that PR.

Testing

• Ran against at least one real endpoint and checked the table renders
  correctly, columns aligned, colors sane
• If this touches the live-printing/column/pointer logic (run_once,
  LIVE_FIELD_KEYS, FINAL_FIELD_KEYS, the failure-marker placement):
  tested a failure path, not just success - e.g. a request that times
  out mid-BODY_DL, a DNS-FAIL, or -6/plain http:// (no TLS phase)
  combined with a failure, since these are exactly the cases where a marker
  has previously gone missing or landed on the wrong column
• Tested with --stream/-S too, if the change touches
  FIELDS/FINAL_FIELD_KEYS (stream mode appends columns at runtime in
  main())
• Checked output with color both on (TTY) and off (piped, e.g. ... | cat),
  if this touches any _col(...)/write_cell/_colorize_* code
• pytest passes
• ruff check . and ruff format --check . pass (see pyproject.toml for
  the configured rule set -
  select = ["E4","E7","E9","F","W","I","UP","B","S"], line length 88,
  double quotes)

contrib/check-endpoint-exporter/check-endpoint.py

• N/A - this PR doesn't touch core request/output logic
• This is a bug fix or behavior change that also applies to the exporter's
  copy of the script, and I've mirrored it there
• This is exporter-specific and intentionally doesn't touch
  check-endpoint.py

(Ruff is configured to skip linting this file a second time since it's a
duplicate of the main script - see extend-exclude in pyproject.toml.)

Anything reviewers should know