parse_value returns string values with their <...> delimiters attached (#176)

[gdevenyi]

Fixes #176.

Spec 2.2: the angle brackets delimit a JCAMP-DX string, they are not part of its value. parse_value returned whole <...> matches, so every consumer had to strip them:

>>> from brukerapi.jcampdx import GenericParameter as G
>>> G.parse_value("<abc>"), G.parse_value("<>"), G.parse_value("<a> <b>")
('abc', '', array(['a', 'b'], dtype='<U1'))

<> reads as the empty string, per point 4 of the issue. A caller still cannot tell ##$KEY=<> from ##$KEY= by value — both are ''.

What changed

• _STRING_RE captures the inside, keeping the two-alternative escape handling from Parse escaped delimiters and nested structs instead of discarding them (J3, J4, J5) #158, and every parse path unquotes. A value whose elements are all delimited strings skips the numeric casts, because <1> <2> is a list of the strings "1" and "2" and never of two numbers.
• Writing stays symmetric. Nothing about a Python str says whether it is a JCAMP string or an undelimited enum symbol, so a parameter keeps whichever form it was already written in. That is what makes read-modify-write — FrameGroupSplitter, SlicePackageSplitter — put <FG_SLICE> and <a.u.> back as they were, while VisuCoreFrameType=MAGNITUDE_IMAGE stays bare.
• The workarounds go: the 26 [1:-1] slices and the delimited literals in brukerapi/config/, and the .strip("<>") / f"<{fg}>" guards in dataset.py, schemas.py, splitters.py and mergers.py.
• Three recipes used [1:-1] to stop #PARAM.method(...) from being read as a #PARAM + accessor substitution; they take parentheses instead ((#ACQ_sw_version).replace(...)).
• Property references, dim_type expectations, docs and the split example follow.

Effect on the public API

dim_type and metadata are the visible ones:

>>> dataset.dim_type
['spatial', 'spatial', 'FG_SLICE', 'FG_DIFFUSION']      # was ['...', '<FG_SLICE>', '<FG_DIFFUSION>']
>>> dataset.metadata['visu_subject']['name']
'CIC_LRMousePhantom'                                     # was '<CIC_LRMousePhantom>'; subj_id already said this

FrameGroupSplitter("FG_ECHO") and filter(parameter='PULPROG', value='RARE.ppg') take the undelimited name; the delimited form no longer matches.

Verification

• Corpus of 3,468 targets: 0 differences. Same load outcome, shape, dtype and error message for every single one, before and after.
• Property references regenerate with the delimiters dropped and nothing else changed — every one of the 171 changed lines is a <X> → X; no affine, date, shape or id moved.
• Write fidelity unchanged: all 143 real visu_pars still rewrite byte-for-byte. (99 reco files differ in where the 78-column wrap falls, identically on master — pre-existing, not from this change.)
• Suite: 2114 passed, 12 skipped. The 5 test_affine_agrees_with_the_paravision_nifti_sform failures are the same ones master has.

🤖 Generated with Claude Code

https://claude.ai/code/session_01NuK1cZi8U54WXAdXMmGpzy

[Copilot]

Copilot was unable to review this pull request because the user who requested the review has reached their quota limit.