feat(assert): assert_between / assert_not_between for numeric ranges

[Chemaclass]

Problem

Numeric comparison covers the open-ended cases — assert_greater_than, assert_greater_or_equal_than, assert_less_than, assert_less_or_equal_than, assert_within_delta — but not the bounded one. Asserting "the duration was between 100 and 500 ms" or "the exit code is in the 4xx range" takes two assertions, which reports two results and two failure messages for one logical check.

Dates already have the bounded form: assert_date_within_range (src/assert/dates.sh). Numbers do not.

Proposal

assert_between <min> <max> <actual> [label]
assert_not_between <min> <max> <actual> [label]

• Inclusive bounds, matching assert_date_within_range — check that assertion first and mirror its semantics and argument order rather than choosing independently.
• Supports integers and decimals, reusing the fixed-point path added for assert_within_delta in perf(assert): compare assert_within_delta in fixed point, not via bc #979 (with the bc/awk fallback) instead of forking bc per call.
• Reports once, not twice — use bashunit::assert_once (src/assert/once.sh) if implemented in terms of other assertions, so the count matches one assertion (Custom asserts: a composed assertion should count and report once #917).

Acceptance criteria

• assert_between 1 10 5 passes; assert_between 1 10 11 fails
• Bounds are inclusive: assert_between 1 10 1 and assert_between 1 10 10 both pass
• Decimals work: assert_between 0.1 0.3 0.2
• Negative numbers work
• min > max is a usage error, not a silently always-failing assertion
• Non-numeric input is a usage error (feat(assert): a missing argument reports a failed assertion, not a usage error #983 / docs(doubles): the spy assertion family disagrees with itself on argument order #984 precedent), not a comparison against empty
• The failure message shows all three values and which bound was violated
• Counts as exactly one assertion (Custom asserts: a composed assertion should count and report once #917)
• assert_not_between is the exact negation and is tested independently
• No bc fork on the common integer path (perf(assert): compare assert_within_delta in fixed point, not via bc #979 precedent) — the fork-budget acceptance tests must still pass

Repo checklist (agent)

• TDD: RED → GREEN → REFACTOR. Use /add-assertion if helpful.
• Bash 3.0+ only: no printf -v, no +=, no declare -A, no [[ ]], no ${var,,}, no &>>.
• Assertions are bare-named (assert_x), helpers are bashunit::-namespaced. Start every assertion with bashunit::assert::should_skip && return 0 and report through bashunit::assertion_failed / bashunit::assertion_passed, matching its neighbours in the same file.
• A missing required argument must go through bashunit::assert::usage_error, not compare against an empty string (feat(assert): a missing argument reports a failed assertion, not a usage error #983).
• Tests: mirror the src layout — tests/unit/assert/<file>_test.sh. There is no assert_fails; test failure output by comparing against print_failed_test, as the existing tests do.
• Docs: add the entry to docs/assertions.md, then regenerate the bashunit doc acceptance snapshot (editing that file breaks it).
• Update the assertion count in README.md ("It ships 73 assertions…" — currently already stale at 74) and add a CHANGELOG.md line under ## Unreleased.
• Gates: make sa, make lint, ./bashunit tests/, ./bashunit --parallel tests/. Never run shfmt -w.
• One issue = one PR.