feat(parser): implement strict indentation mode (LogseqIndentationError)

[MarcoPorcellato]

Problem

LogseqIndentationError is defined and tested for hierarchy placement (tests/test_exceptions.py) but is never raised by StackMachineParser. Misaligned bullets are silently floored to the nearest indent level (test_misaligned_indentation_floors_to_nearest_level).

Operators and integrators cannot opt into fail-fast parsing for corrupted vault indentation.

Proposed Solution

• Add strict_indent: bool = False to StackMachineParser (or wire LogseqIndentationError to an existing strict flag)
• When enabled, raise LogseqIndentationError on non-monotonic indent jumps that today are coerced
• Document behavior in ARCHITECTURE.md and logseq_ast_primer.md
• Table-driven tests in tests/test_logos_parser.py

Context

Reserved exception surfaced during local code study (wave 9). Aligns with docs/roadmaps/ROADMAP_ROBUSTNESS_AND_SOFT_BREAKS.md tension between lenient vs strict modes.

Out of scope

• Changing default lenient behavior (breaking change)

Definition of Done

• Opt-in strict mode with tests
• make all passes
• CHANGELOG.md entry under [Unreleased] if behavior is user-visible