Reflection table improvements

[dimitrivlachos]

This PR works through the open items collected in #15, which came out
of using the DX2 ReflectionTable in the fast-feedback-service indexing
algorithm. Each was a small gap in the table's public API or a
surprising write-time behaviour; they are addressed as three
self-contained commits.

Changes:

• Add a method to return the reflection table size #20
  Add ReflectionTable::size(). Exposes the number of rows
  (reflections) as a public method (include/dx2/reflection.hpp,
  dx2/reflection.cxx), delegating to the existing private
  get_row_count() helper. Previously callers had to fetch an arbitrary
  column and read its extent(0) just to learn the row count.

• Reflection table writing appears to update an existing file #19
  Truncate the output file on write().
  ReflectionTable::write now opens the target with
  H5Fcreate/H5F_ACC_TRUNC (dx2/reflection.cxx). It previously
  opened an existing file with H5F_ACC_RDWR and only overwrote
  same-named datasets in place, so datasets from a prior write that were
  absent from the current table were left behind as stale data (it did
  not append rows, as first suspected). Truncating guarantees the file
  contains only the current table's columns.

• Allow reassignment of reflection column data? #18
  Add ReflectionTable::update_column. Adds a strict
  update_column<T> with shape-based, 1D and 2D overloads
  (include/dx2/reflection.hpp) mirroring add_column. It requires the
  named column to already exist (throwing otherwise), validates type
  support and row-count consistency with the rest of the table, and
  replaces the column in place to preserve column order. add_column
  keeps its duplicate-name safety net, so the previous "check if
  present, else add" workaround is no longer needed. The shared
  type-support check is factored into a private
  ensure_supported_type<T>() helper reused by both methods.

Each change is covered by new gtest cases in tests/test_reflection_table.cxx
(size of empty/populated/loaded tables, overwrite-not-update on
re-write, and update_column replace / missing / row-count-mismatch /
order-preservation).

Closes #15.
Closes #18.
Closes #19.
Closes #20.

[jbeilstenedmands]

Thanks for addressing these issues and adding clear tests for the expected functionality. I have also tested the new features work as expected.