RAMTools 0.1.0

The first tagged release. RAMTools began in 2017 at CERN openlab as a format
stored in ROOT's TTree; this release is its rewrite on RNTuple. Files written by
it stay readable by every later release.

Added

• The RAM format on ROOT's RNTuple and samtoramntuple, which converts SAM to
  it (#1, #2).
• bamtoramntuple, which converts BAM through htslib (#20).
• ramdump, which writes a file or a region of it back out as SAM and takes the
  samtools view options -h, -H, -c, -f, -F and -o (#75).
• ramntupleview, which counts the records in a region and reports the query
  time (#10, #69).
• -split, one output file per reference, written as a stream and indexed like
  any other file (#9, #79).
• -compression N on both converters, taking a ROOT compression code (#73).
• -threads N on both converters; samtoramntuple parses, encodes and
  compresses 64 MB blocks on every thread and writes them in input order
  (#86, #87, #88).
• Region queries find the region by binary search over the positions of a
  coordinate-sorted file; the converters record whether the input was sorted,
  and an unsorted file is read in full (#79, #83).
• Benchmarks: a SAM generator and Google Benchmark suites for conversion and
  region queries (#4, #6, #7, #8), and scripts that run them on real data
  (#54).
• Continuous integration: build and tests, clang-tidy review, formatting checks
  and Codecov gating (#3, #5, #11, #17, #18, #19, #21, #32, #35,
  #47, #57, #77).
• Tests for the index, the quality encodings and the record getters (#39,
  #41, #43).
• The Apache License 2.0 (#60); CONTRIBUTING.md, issue and pull request
  templates, and a README that walks through installation, conversion and
  queries (#12, #89).

Changed

• ROOT 6.38 is the minimum version (#76).
• SEQ is stored as uppercase text instead of 4-bit packed bases, which
  compresses better: 3.9% smaller files on chromosome 11. Files with packed
  sequences still read (#85).
• Opening a RAM file no longer starts ROOT's interpreter. On 2 million records
  of chromosome 11, counting a 10 Mb region went from 0.59 s and 150 MB to
  0.26 s and 19 MB (#92), and dumping a 1 kb region from 0.61 s and 204 MB to
  0.30 s and 78 MB (#93). -h and -H still start it, to read the header.
• Region queries decide overlap by each read's CIGAR span and accept rname,
  rname:pos, rname:start-end and * (#15).

Removed

• The TTree backend (#71).
• RAMNTupleConverter, replaced by ramdump (#75).
• The sparse index of mapped reads (#16), replaced by the binary search
  (#83).

Fixed

• Sequences decoded wrongly by DecodeSequence (#45).
• The SAM parser accepted non-numeric and out-of-range integer fields, and a
  query for an unknown reference added it to the reference list (#59).
• Float tags lost precision (#64).
• A CIGAR of * came back as an empty field (#66).
• Illumina quality binning was indexed by the ASCII character instead of the
  Phred score (#68).
• Fields lost or shifted on the way in or out: lines over 10 kB, consecutive
  tabs, unknown and lowercase bases, unknown CIGAR operators (#70).
• The SAM header was written as one key per line and could not be read back
  (#72).
• A mapped read with a malformed CIGAR such as 10Q was stored without one; it
  is now reported with its line number and skipped (#74).
• A region query past the last index anchor of a reference started from the
  first row and read the whole file (#82); the index was later replaced by the
  binary search (#83).
• README examples that did not run (#78).
• bamtoramntuple did not link when htslib is not in a default library
  directory, as on macOS (#95).
• clang-tidy findings and suppressions (#14, #38).

Full list of changes: CHANGELOG.md.