docs: clarify retry-max-time timing check and sleep behavior

[dioput12]

This change clarifies the documentation for --retry-max-time when used together with --retry and --retry-delay.

The existing text already states that retries continue “as long as the timer has not reached this given limit”, but it does not explicitly mention that the retry timer includes the sleep time between retries (for example due to --retry-delay) and that the limit is checked before starting each new retry.

The new paragraphs make this behavior explicit and also restate that transfers which have already started are allowed to complete even if this makes the total wall clock time exceed the limit, with --max-time available to cap the duration of each individual transfer attempt.

[dioput12]

My testing script:

#!/usr/bin/env bash
# retry-max-time regression notes for example.invalid
# Run from a shell with curl in PATH.

TARGET="http://example.invalid"

echo "== Baseline: retry-max-time 5 =="
time curl -v \
  --retry 3 \
  --retry-delay 2 \
  --retry-max-time 5 \
  --connect-timeout 2 \
  "$TARGET"

echo
echo "== retry-max-time 3 (expect fewer retries than 5s case) =="
time curl -v \
  --retry 3 \
  --retry-delay 2 \
  --retry-max-time 3 \
  --connect-timeout 2 \
  "$TARGET"

echo
echo "== retry-max-time 1 (only initial attempt + at most one retry) =="
time curl -v \
  --retry 3 \
  --retry-delay 2 \
  --retry-max-time 1 \
  --connect-timeout 2 \
  "$TARGET"

echo
echo "== retry-max-time 10 with max-time 4 (per-attempt limit) =="
time curl -v \
  --retry 3 \
  --retry-delay 2 \
  --retry-max-time 10 \
  --max-time 4 \
  "$TARGET"

[bagder]

Thanks!