Treat a blank at the start of a value block as layout, not as value

[gdevenyi]

Fixes #193.

_normalize_line_breaks only rstrip()ed, so the blank ParaVision writes after an opening ( stayed glued to element 0. parse_value tests the string shape with val_str.startswith("<"), which is then False, so the only branch that calls unquote_string is skipped and the token is returned verbatim:

parse_value('( <A>, <B> )')  ->  ['<A>', 'B']      # asymmetric
parse_value('( 1, 2 )')      ->  ['1', 2]          # leading int becomes a str

On real files:

##$SUBJECT_study_adj_config=( <MRI_Default> , <MRI Default> )
  ->  ['<MRI_Default>', 'MRI Default']

so value[0] == 'MRI_Default' was silently False, and the write path re-quoted the kept delimiters into <<MRI_Default>>. 613 parameter occurrences across the corpus were affected — CONFIG_SCAN_operation_mode (the spec's own worked example for this shape), AdjConfigurationMode, CONFIG_SCAN_RxCoilsNames, CONFIG_SCAN_TxCoilsNames, SHIM_status_check_sum_components, SUBJECT_study_adj_config, PVM_StartupShimList.

Change

rstrip() → strip(), at the one shared point. That repairs the string case and the numeric case together. A value block begins after ##$name= or after a ( / , delimiter and ends where the next ## record begins, so blanks at either end are layout. Blanks inside the value are untouched — ParaVision's wrap puts the space before the newline it inserts, so joining the lines still reassembles the string.

After the fix, on real files:

SUBJECT_study_adj_config   -> ['MRI_Default', 'MRI Default']
PVM_StartupShimList        -> ['S0', 'Spherical Shim.']
CONFIG_SCAN_RxCoilsNames   -> ['1', 'Chan1[1H]: RF ARR 400 1H M.BR. 2x2 RO AD']

Tests

• test_a_blank_after_the_opening_paren_is_layout_not_value — the three shapes plus a subject file written inline. Fails on master.
• test_blanks_inside_a_wrapped_string_are_kept — a wrap inside a <...> string, guarding the blanks the change must not touch.

Suite: 2140 passed, 12 skipped (unchanged). Corpus load test unchanged at 3202/3478 with an identical failure breakdown.

[gdevenyi]

Rebased onto #213 (ruff format), so this branch carries that commit until #213 merges — merge #213 first and this diff reduces to its own change. Verified after the rebase: ruff check and ruff format --check clean, full suite green.

Integration check across all eleven conformance PRs merged together: 2164 passed, 12 skipped; corpus load test 3202 → 3207 of 3478 with zero newly broken datasets.