docs: add literature references throughout the code (closes #3)

[jasasonc]

Closes #3.

Currently the references live only in the README. This adds the literature citations into the code so the theory behind each formula is discoverable at the point of use.

What was added

• Module-level bibliography in spectrum.py and tools.py — Lalanne, Mechanical Vibration and Shock Analysis, Vol. 1/3/4/5 (2009, ISTE/Wiley) and Thomson, with consistent [LalanneN]/[Thomson] reference keys.
• Spectrum class docstring mapping each signal type (sine / sine-sweep / random-PSD / random-time) to its governing equation.
• Per-method and inline references with exact equation numbers:
  • sine ERS → eq [1.7], FDS → eq [3.19]
  • sine-sweep ERS → eqs [1.31]–[1.33], FDS → eq [3.65] + dwell weighting Table 3.2
  • random-PSD: segmented response RMS → eq [8.86], I0/I2/I4 closed forms → App. A6 ([A6.20]/[A6.22]/[A6.24]), n0 → eq [5.76], FDS → eq [4.41]/[4.7]–[4.8]
  • random-time FDS → eq [4.6]; damped SDOF impulse response (Thomson)

Corrections to pre-existing citations

• Sine-sweep ERS comments cited Vol. 1 "[1.21]/[1.22]/[1.23], p.31-32" — the correct source is Vol. 5 eqs [1.31]/[1.32]/[1.33], p.16 (the below-/above-band tags were also swapped).
• Random-PSD FDS cited "Vol. 0 … (A1-93)" → replaced with Vol. 4 eq [4.41] = Vol. 5 eqs [4.7]–[4.8].
• response_relative_displacement cited Thomson eq (4.2-5), which is the undamped kernel; clarified that the code implements the damped impulse response.

Notes

• Documentation only — no behaviour change. Full test suite passes (10 passed).
• Stacked on fix/random-fds-factor-2, so the random-FDS reference comments describe the corrected n0/full-cycle formulas. Base this PR on that branch (or rebase onto master once the fix lands).